<%BANNER%>

A Human Factors Approach to File Storage and Retrieval

Permanent Link: http://ufdc.ufl.edu/UFE0021242/00001

Material Information

Title: A Human Factors Approach to File Storage and Retrieval
Physical Description: 1 online resource (135 p.)
Language: english
Creator: Alfano, Keith Michael
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2007

Subjects

Subjects / Keywords: cognitive, file, memory, psychology, retrieval, search
Psychology -- Dissertations, Academic -- UF
Genre: Psychology thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: The question of whether people can directly retrieve or must compute an answer or solution to a question or problem has been studied in a number of problem-solving domains. The purpose of the present research was to extend this question into the realm of file retrieval on the personal computer (PC). Which cognitive strategy, memory or search, contributes most heavily to navigation decisions made by PC users as they seek for target files? In Experiment 1, the nodes of two seven-level, balanced binary trees were populated with folder and file names, such that one tree made logical, hierarchical sense, while the other was nonsensical. In Experiment 2, participants stored and retrieved files in those trees, using exclusively memory, exclusively search, or a combination thereof, depending on the condition to which they were randomly assigned. While the speed of search-based traversal decisions increased as function of tree depth, memory speeds displayed no such distance-to-goal sensitivity. Conversely, the accuracy of search-based traversal decisions was consistently high across tree levels, whereas the accuracy of memory was deplorable, particularly at deeper tree levels. Interestingly, when both memory and search were available, traversal decisions were both faster and more accurate than those made when only one of the two cognitive strategies was available. These findings can, in part, be explained in terms of a choice model and/or race-horse model.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Keith Michael Alfano.
Thesis: Thesis (Ph.D.)--University of Florida, 2007.
Local: Adviser: Delaney, Peter F.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2009-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2007
System ID: UFE0021242:00001

Permanent Link: http://ufdc.ufl.edu/UFE0021242/00001

Material Information

Title: A Human Factors Approach to File Storage and Retrieval
Physical Description: 1 online resource (135 p.)
Language: english
Creator: Alfano, Keith Michael
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2007

Subjects

Subjects / Keywords: cognitive, file, memory, psychology, retrieval, search
Psychology -- Dissertations, Academic -- UF
Genre: Psychology thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: The question of whether people can directly retrieve or must compute an answer or solution to a question or problem has been studied in a number of problem-solving domains. The purpose of the present research was to extend this question into the realm of file retrieval on the personal computer (PC). Which cognitive strategy, memory or search, contributes most heavily to navigation decisions made by PC users as they seek for target files? In Experiment 1, the nodes of two seven-level, balanced binary trees were populated with folder and file names, such that one tree made logical, hierarchical sense, while the other was nonsensical. In Experiment 2, participants stored and retrieved files in those trees, using exclusively memory, exclusively search, or a combination thereof, depending on the condition to which they were randomly assigned. While the speed of search-based traversal decisions increased as function of tree depth, memory speeds displayed no such distance-to-goal sensitivity. Conversely, the accuracy of search-based traversal decisions was consistently high across tree levels, whereas the accuracy of memory was deplorable, particularly at deeper tree levels. Interestingly, when both memory and search were available, traversal decisions were both faster and more accurate than those made when only one of the two cognitive strategies was available. These findings can, in part, be explained in terms of a choice model and/or race-horse model.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Keith Michael Alfano.
Thesis: Thesis (Ph.D.)--University of Florida, 2007.
Local: Adviser: Delaney, Peter F.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2009-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2007
System ID: UFE0021242:00001


This item has the following downloads:


Full Text

PAGE 1

1 A HUMAN FACTORS APPROACH TO FILE STORAGE AND RETRIEVAL By KEITH M. ALFANO A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2007

PAGE 2

2 200 7 Keith M. Alfano

PAGE 3

3 ACKNOWLEDGMENTS I wish to convey my appreciation to my dissertation advisor and committee chairperson, Dr. Peter Delaney, for his invaluable contributions to the compilation of this manuscript. I would also like to thank my current and former committee members: Drs. Lise Abrams, Scott Miller, Joseph Wilson, and Paul Fishwick, as well as my research assistant, Stefine DeLisser, for their contributions to the project. The lab space a nd human resources needed to conduct this research were provided by the Psychology Department at the University of Florida and are much appreciated. Finally, I would like to thank my parents for their never ending support, without which this project would not have been possible.

PAGE 4

4 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 3 LIST OF TABLES ................................ ................................ ................................ ........................... 6 LIST OF FIGURES ................................ ................................ ................................ ......................... 7 ABSTRACT ................................ ................................ ................................ ................................ ..... 9 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 10 Problem Solving in Information Space ................................ ................................ ................... 11 Human Behavior in Information Spaces ................................ ................................ ................. 19 Goals of Users ................................ ................................ ................................ ................. 19 Cognitive Maps ................................ ................................ ................................ ............... 20 Cognitive Models of Search in Familiar Information Spaces ................................ ................ 21 Episodic Memory Versus Search D uring Information Search ................................ ............... 28 Summary and Conclusions ................................ ................................ ................................ ..... 37 Present Research ................................ ................................ ................................ ..................... 38 Rationale ................................ ................................ ................................ .......................... 38 Experimental Objectives ................................ ................................ ................................ 39 2 MATERIALS AND METHODS ................................ ................................ ........................... 41 Experiment 1 ................................ ................................ ................................ ........................... 41 Participants ................................ ................................ ................................ ...................... 41 Design and Procedure ................................ ................................ ................................ ...... 41 Experiment 2 ................................ ................................ ................................ ........................... 43 Participants ................................ ................................ ................................ ...................... 43 Design and Procedure ................................ ................................ ................................ ...... 43 Me mory condition (baseline) ................................ ................................ ................... 45 Search condition (baseline) ................................ ................................ ...................... 46 Memory/Search match condition (experimental) ................................ .................... 47 Memory/Search no match condition (experimental) ................................ ............... 47 Summary of file retrieval conditions ................................ ................................ ........ 48 Binary Tree Specifications ................................ ................................ .............................. 49 3 RESULTS AND DISCUSSION ................................ ................................ ............................. 58 Experiment 1 ................................ ................................ ................................ ........................... 58 Experiment 2 ................................ ................................ ................................ ........................... 59 Overall Findings by Condition: Speed and Accuracy of Traversal Decisions ............... 60 els ................................ ................................ ................................ ...... 63

PAGE 5

5 Planning time (level 0) ................................ ................................ ............................. 63 Leaf nodes (Level 7) ................................ ................................ ................................ 64 Level 6 ................................ ................................ ................................ ...................... 64 Speed of Traversal Decisions: Condition by Level Interaction ................................ ...... 65 Baseline findings ................................ ................................ ................................ ...... 66 Experimental findings ................................ ................................ .............................. 67 Accuracy of Traversal Decisions: Condition by Level Interaction ................................ 68 Baseline findings ................................ ................................ ................................ ...... 69 Experimental findings ................................ ................................ .............................. 70 Relating Speed and Accuracy of Node to Node Traversal Decisions ............................ 71 Task Difficulty Measures ................................ ................................ ................................ 72 Item Analyses ................................ ................................ ................................ .................. 74 Practice Effects ................................ ................................ ................................ ................ 75 Individual Differences in Working Memory Capacity and Storage Retrieval Time Interval ................................ ................................ ................................ ......................... 77 Subjective Reports of Strategy ................................ ................................ ........................ 80 Semantic Organization of Folder Names ................................ ................................ ........ 82 4 GENERAL DISCUSSION ................................ ................................ ................................ ... 101 Summary of Findings ................................ ................................ ................................ ........... 102 Population of Binary Trees (Experiment 1) ................................ ................................ .. 102 Speed of Memory and Search Baseline Traversal Decisions (Experiment 2) ............... 103 Accuracy of Memory and Search Baseline Traversal Decisions (Experiment 2) ......... 104 Memory/Search Hybrid Performance (Experiment 2) ................................ .................. 105 Speed Accuracy Relationships in Traversal Decisions ................................ ................. 107 Spatial, Symbolic, and Working Memory ................................ ................................ ..... 108 Implications and Applications ................................ ................................ .............................. 110 Ecological Validity ................................ ................................ ................................ ........ 110 PC Based File/Folder Exploration Tools ................................ ................................ ...... 112 Future Work ................................ ................................ ................................ .......................... 114 APPENDIX A EXPERIMENT 2: TREE A FOLDER/FILE NAMES ................................ ........................ 118 B EXPERIMENT 2: TREE B FOLDER/FILE NAMES ................................ ........................ 124 LIST OF REFERENCES ................................ ................................ ................................ ............. 130 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ....... 135

PAGE 6

6 LIST OF TABLES Table page 2 1 Semantic relatedness and super/sub rating scales for Experiment 1. ................................ 51 2 2 Summar y of file retrieval conditions for Experiment 2. ................................ .................... 53 2 3 Stored and retrieved file names, descriptions, and locations for all four conditions. ........ 54 2 4 Stored and retrieved file names, descriptions, and locations for all four conditions. ........ 55 3 1 Mean Relatedness and Super/Sub scores and inclusion criteria for Tree A and Tr ee B folder names. ................................ ................................ ................................ ...................... 83 3 2 Log TON P(BACK) correlations, overall and by condition. ................................ ............. 85 3 3 Performance metrics and values for OSPA N, storage, and retrieval by condition. ........... 85 3 4 Correlation matrix showing storage and retrieval metrics across all conditions ............... 86 3 5 Correlation matrix showing storage and retrieval metrics for Memory condition. ........... 86 3 6 Correlation matrix showing storage and retrieval metrics for Search condition. .............. 87 3 7 Correlation matrix showing storage and retrieval metrics for Memory/Search match condition. ................................ ................................ ................................ ........................... 87 3 8 Correlation matrix showing storage and retri eval metrics for Memory/Search no match condition. ................................ ................................ ................................ ................. 88 3 9 Names, descriptions, locations and mean performances associated with specific files. .... 89 3 10 Correlation matrix showing serial position of file and performance measures. ................ 93 3 11 Correlation matrix showing OSPAN score, storage retrieval lag time, and performance measures (overall and by condition). ................................ ............................ 94 3 12 Correlation matrix showing self reported strategies and performance measures and participant sex. ................................ ................................ ................................ ................... 96 3 13 Correlation matrix showing folder name ratings and performance measures by tree. ...... 99 4 1 Potential traversal speed findings and their theoretical interpretations (Exp. 2) ............ 116 4 2 Potential traversal accuracy findings and their theoretical interpretations (Exp. 2). ....... 116

PAGE 7

7 LIST OF FIGURES Figure page 2 1 Portion of Tree A. ................................ ................................ ................................ .............. 51 2 2 Portion of Tree B. ................................ ................................ ................................ .............. 51 2 3 Top four levels of balanced binary tree ................................ ................................ ............. 52 2 4 Storage phase user interface with navigation controls (practice mode). ........................... 52 2 5 Retrieval phase user interface with navigation controls (practice). ................................ ... 53 2 6 Binary tree coordinate system used for node identification ................................ ............... 54 3 1 Comparison of nod e based decision times by condition. ................................ .................. 83 3 2 Comparison of backup frequencies by condition. ................................ .............................. 83 3 3 Comparison of node based decis ion times by condition and tree level. ............................ 84 3 4 Comparison of backup frequencies by condition and tree level. ................................ ....... 84 3 5 Scatter plots and regression lines for log file serial position versus task performance measures. ................................ ................................ ................................ ............................ 93 3 6 Scatter plots and regression lines for log OSPAN score versus task perfo rmance measures ................................ ................................ ................................ ............................ 95 3 7 Scatter plots and regression lines for storage retrieval lag time and task performance measure s ................................ ................................ ................................ ............................ 95 3 8 Double dissociation between self reported memory strategy and sex. .............................. 96 3 9 Scatter plots and regression lines for self reported overall strategy (Memory %) and file storage pe rformance measures ................................ ................................ .................... 97 3 10 Scatter plots and regression lines for self reported overall strategy (Memory %) and file retrieval perfor mance measures ................................ ................................ .................. 97 3 11 Scatter plots an d regression lines for self reported memory strategy (Spatial %) and file storage per formance measures ................................ ................................ .................... 98 3 12 Scatter plots and regression lines for self reported memory strategy (Spatial %) and file retrieval performance measur es ................................ ................................ .................. 98 3 13 Scatter plots and regression lines for mean TON versus folder name pair score ............. 99

PAGE 8

8 3 14 Scatter plots and regression lines for mean P(BACK) versus folder name pair score ... 100 4 1 on task. ......................... 117

PAGE 9

9 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy A HUMAN FACTORS APPROACH TO FILE STORAGE AND RETRIEVAL By Keith M. Alfano August 2007 Chair: Peter F. Delaney Major: Psychology The question of whether people can directly retrieve or must compute an answer or solution to a question or problem has been studied in a number of problem solving domains. The pu rpose of the present research was to extend this question into the realm of file retrieval on the personal computer (PC). Which cognitive strategy memory or search contributes most heavily to navigation decisions made by PC users as they seek for target f iles? In Experiment 1, the nodes of two seven level, balanced binary trees were populated with folder and file names, such that one tree made logical, hierarchical sense, while the other was nonsensical. In Experiment 2, participants stored and retrieve d files in those trees, using exclusively memory, exclusively search, or a combination thereof, depending on the condition to which they were randomly assigned. While the speed of search based traversal decisions increased as function of tree depth, memor y speeds displayed no such distance to goal sensitivity. Conversely, the accuracy of search based traversal decisions was consistently high across tree levels, whereas the accuracy of memory was deplorable, particularly at deeper tree levels. Interesting ly, when both memory and search were available, traversal decisions were both faster and more accurate than those made when only one of the two cognitive strategies was available. These findings can, in part, be explained in terms of a choice model and/or race horse model.

PAGE 10

10 CHAPTER 1 INTRODUCTION The amount of information with which human beings interact has exploded in recent years. Therefore, the question of how individuals search, process, and retrieve information has become increasingly important. As use of personal computers (PCs) and the Internet become ubiquitously widespread, the ability to explore and search virtual information spaces has become a particularly interesting area to study within the field of human computer interaction (HCI). The qu estion of what cognitive processes underlie decision making during an information search task is extremely interesting, with a plethora of potential implications for applied software development. For example, if the cognitive processes by which people sea rch for and retrieve files on their PCs are better understood, software developers could exploit these human factors. The file explorer applications that reside within future operating systems could be designed to expedite human search for files and direc tories of interest. This paper serves two major functions: 1) to review the literature, including relevant cognitive models, on human exploration and navigation of virtual information space and 2) to explore the relative contributions of search and memory in the decision making processes associated with searching virtual information environments. First, search tasks conducted within virtual information spaces are discussed in terms of a traditional problem solving and problem space paradigm. It is argued that traversal of information space is a process analogous, if not identical, to the process of problem solving, an area germane to traditional cognitive psychology. Next, a discussion of the potentially different, but not mutually exclusive, goals and b ehaviors exhibited by users while navigating through virtual information space ensues. An overview of some of the HCI cognitive engineering models associated with virtual search in familiar information spaces is then provided. Models which incorporate an episodic memory component

PAGE 11

11 into their architectures are emphasized. The paper concludes with a discussion about the relative contributions of episodic memory and more effortful search while individuals search for targets in virtual information space. Thi s description provides a framework for the present study. Problem Solving in Information Space A major assumption of this paper is that traversal of information space is analogous, if not identical, to problem solving processes. That is, each node within an information space can be viewed as equivalent to a problem state. Just like decision making during problem solving, traversal direction decisions in information space may lead either toward dead ends or toward the goal state. Perhaps there is just one difference between problem solving in information space more generally. Beginning with the work of Newell and Simon (1972), it has been customary to analyze problem solving tasks in the context of problems space. The problem space consists of the following components: 1) an initial state, 2) a goal state that is to be achieved, 3) operators for transforming the problem from the initial state to the goal state in a sequence of steps, and 4) constraints on application of the operators that must be satisfied. The problem solving process itself (i.e., retrieving a ta rget from within a hierarchically structured information space) may be conceived as a search for a path that connects the initial and goal states. Therefore, by this definition, if a problem solving task resides within a virtual information space, it may be possible to map the problem solving space directly on to the information space, thus forming an information based problem space. In this case, the various information states, or nodes, would become one and the same with the problem states. A user may traverse these states, or nodes, according to some predefined structure that links them.

PAGE 12

12 In ordinary problem spaces, it may be difficult to operationally define the point at which a participant is currently residing at a given state. However, in informati on spaces, a state is much easier to define, given that it is synonymous with the node in information space at which the user currently resides. In addition to helping experimenters and system designers operationally define states, a structured informatio n space may help a user reach a desired node where a target object is located. In problem solving terminology, this would refer to problem solution. It has been shown that problems consisting of well defined problem states are easier to solve than proble ms with ill defined states. For example, an actual, tangible, hand controlled Chinese Ring Puzzle was considerably more difficult to solve than its virtual, computer based Balls and Boxes isomorph ( Kotovsky & Simon, 1990). In fact, while participants wer e unable to solve the Chinese Ring Puzzle after more than two hours of fiddling with it, participants working on the Balls and Boxes isomorph were usually able to achieve problem solution (i.e., the goal state) in under twenty minutes. Not surprisingly, what constitutes a problem state and, hence, the criteria for whether or not a participant has changed states, in the real world Chinese Ring Puzzle is much more poorly defined than what constitutes a problem state in the Balls and Boxes isomorph. In shor t, a problem space, or an information space, consisting of well defined states is beneficial to experimenters (or system designers) and participants (or end users), alike. While helping designers and experimenters to operationally define states and depend ent variables, it helps users and participants to achieve problem solution. In addition to the clarity of problem states (or nodes within information space), other issues dealing with surface representations of a problem may influence its level of difficul ty. Researchers have found large differences in difficulty and varying amounts of transfer among isomorphs of the Tower of Hanoi, for example (Kotovsky, Hayes, & Simon, 1985). Since the

PAGE 13

13 tasks have the same formal structure, these differences must result from surface representations of the problems. Using verbal protocols, Kotovsky and colleagues rationalized these differences by describing the various degrees to which a set of rules may relate to real world knowledge. In addition, the rules for some iso morph may be externalized to a greater extent than others. That is, the rules may be embedded in the external problem representation and may even, at times, appear obvious (Zhang, 1997). Gunzelmann and Blessing (2000) presented participants with twelve p roblems for each of three isomorphs of the Tower of Hanoi (the Standard Tower of Hanoi, Monster/Globe Move, and Paint Stripping). Results showed that any of the isomorphs were sufficient to produce transfer. More interestingly, the comparison of tasks sh owed that the Monster/Globe Move isomorph was most difficult, followed by the Paint Stripping isomorph, with the standard Tower of Hanoi being the easiest of the three tasks. While the rules for the Tower of Hanoi were largely inferable from the presentat ion, all of the Monster/Globe Move rules needed to be learned explicitly. Furthermore, at least some of the Paint Stripping rules were not intuitive based solely on its presentation. These results suggest that incorporation of problem constraints, or rul es, into the problem representation can reduce cognitive load, thereby reducing the level of problem difficulty. The size of a problem space increases exponentially with the complexity of a problem (Proctor and Vu, 2003). This fact has two important impli interchangeably with difficulty. However, failure to make a distinction may be a source of confusion. A task can be difficult beca use it requires a great deal of cognitive effort, without necessarily being complex. On the other hand, a task can be difficult because of the high degree of skill or knowledge that is required (Campbell, 1998). Second, this exponential growth often

PAGE 14

14 mean s that problem spaces are well beyond the capacity of short term episodic memory. Therefore, users must employ some degree of effortful search in order to facilitate intermediary decision processes and, ultimately, achieve the goal state (e.g., locate the target node). Means end analyses and hill climbing techniques are general paradigms and heuristics used in search. They are generally considered to be weak methods, as they do not use domain knowledge during problem solving. Means end analysis is a gene ral paradigm which attempts to minimize differences between the current state and the goal. It chooses a difference and then applies an operator to remove. If the operator cannot be applied directly, a sub goal is established to reach a state in which di rect application is possible. This type of backward chaining is known as operator sub goaling and makes the means ends heuristic somewhat stronger and more sophisticated than hill climbing (Dunbar, 1998; Lovett, 2002). Newell, Shaw, and Simon (1957) cre ated a General Problem Solver (GPS) model, in which each rule had by the differences they could reduce. Hill climbing occurs when the problem solver choos es to apply an operator that will lead to a state that is closer to the goal than that obtained by applying current situation (Lovett, 2002). Hill climbing, like all heuristics, does not guarantee problem solution. Its major shortcoming is its assumption that local improvement will lead to global improvement. Its specific problems include local maxima (reaching the top of the hill but missing the mountain), plat eaus (lacking motivation to move anywhere because everything around appears equally good to the current state), ridges (needing to apply multiple operators, instead of just a single operator, in order to advance toward problem solution on an upward slope). Therefore, under certain circumstances, other search methods and techniques may be

PAGE 15

15 preferable (e.g., backtracking, making big jumps, applying multiple rules before testing). Hill climbing is best suited to problems where the heuristic gradually improves the closer it gets to the solution. It works poorly where there are sharp drop offs. Nearly thirty years ago, Atwood and Polson (1976) developed a process model for the water jug problem solving task. The model assumed means ends analysis and hill climb ing move selection heuristics. Atwood and Polson suggested that all previous move positions are committed to long term memory (LTM) and that this allows problem solvers to make effective use of an anti looping heuristic. That is, memory of previous posit ions would enable problem solvers to avoid previously visited states. Models such as these imply that strategies unfold in a temporally forward direction, that problem solvers typically restrict forward planning activities to just one or two moves ahead o f the current problem state, and that familiarity is a heuristic used in the avoidance of previous moves. Two years earlier, Reed, Ernst, and Banerji (1974) argued that memory alone does not regulate transfer of cognitive skill between the same or similar problem states. Using two isomorphic problems [the Jealous Husbands (JH) problem and the Missionary Cannibal (MC) problem], they studied the transfer of cognitive skill at the level of individual moves, general strategies, and practice in applying operato rs. The number of illegal moves was reduced for both problems. However, the authors failed to find significant transfer in the MC JH direction despite the fact that participants remembered a good deal about how they solved the problem, as indicated by th eir reduction in solution time when asked to solve the same problem again. Only in the JH MC direction and when participants were informed of the relationship between the two isomorphs was significant transfer observed. These findings suggested that abse nce of transfer may be a product of failure to recognize or make use of analogy, as opposed to a memory failure.

PAGE 16

16 Davies (2000) also argued that the transfer of cognitive skill (i.e., the anti looping heuristic) may be oversimplified in some older models Move estimation tasks may not directly different from recognition of actual problem states encountered by the problem solver. Davies contended that the heuris tic may involve participants planning retrospectively in order to reconstruct previous move information. Using the Tower of Hanoi task, Davies found that estimation accuracy and estimation time for a particular move position were both systematically relat ed to the distance from the current state when the state is presented to participants. When the current state was absent, then estimation accuracy decreased, along with estimation time and the accuracy of recognition memory for previous states of the prob lem. These data suggested that provision of current state information may enable participants to use it as a basis for retrospective planning. In addition, this study demonstrated that estimations about where future positions are likely to occur were sym metrical to estimations about past positions, as measured by times required to make judgments and distances from the current state. This suggests that backwards planning may share some fundamental characteristics with forward planning processes. For exam ple, for both types of planning processes, recognition accuracy (and times) and distance from the current problem state to the position that the participant is asked to recognize or estimate share a linear relationship. Retrospective planning may be seen as similar to the backup operator applied by AYN, a function model developed by Howes (1994). This backup operator is applied by AYN under certain conditions to constrain search space (see the section on Cognitive Models of Search in Familiar Information Spaces). Although backwards planning may interact with memory processes, memory processes alone probably do not account for the anti looping heuristic adopted by problem solvers (Davies,

PAGE 17

17 2000). Intuitively, this makes sense. It is clearly inefficient to remember all previous problem states. For example, in most well structured problems (e.g., water jugs, Tower of Hanoi), problem states that are distant from the current state would be of little relevance, as they possess a significantly lower probability of re entry. It may be possible that short term memory systems may collaborate with retrospective planning processes to recreate previous problem states. However, given the well known capacity limitations of such systems (e.g., Hitch & Baddeley, 1976), t his would only permit limited consideration of previous moves. Problem solvers often have multiple operators available to them and must select only one of them to apply to the task at hand. As evidenced in the discussion above, there is some degree of ten sion in the literature between models of problem solving that emphasize experience and familiarity and those that address current context and concurrent problem solving processes. Lovett and Anderson (1996) sought to reconcile these two schools of thought using the building sticks task (BST), in which participants were to add and subtract the lengths of building sticks to create a current stick that was equal in length to the desired stick. Lovett and Anderson determined that problem solvers used the foll owing two sources of information to make operator selections: 1) history of their success (or lack thereof) when applying certain operators and 2) information from the current context of the problem (heuristics based on the current context). Solvers were more likely to apply operators 1) with which they had prior success and 2) by which the current state would be moved closer to the goal state. First, because of the length of the initial building stick provided by the experimenters, participants procured a gradually increasing history of success of an overshoot strategy during the BST, beginning with the training phase. Overshoot was instantiated by selecting a building stick that is longer than the desired stick and adding it to the current stick of len gth zero. Following an application of

PAGE 18

18 overshoot, solvers selected among operators that subtract from the current stick. Operator selection data showed that problem solvers exhibited tendencies to employ overshoot strategies for both the remainder of that phase and all subsequent phases. Second, current problem features, specifically the distance to goal information that they conveyed, also played a significant role in influencing operator selection. Problem bias, defined as the difference between the hi ll climbing distances for overshoot and undershoot, accounted for nearly 50% of BST model predicted that latencies should decrease as a power law of practice with each production. However, excellent power curve fits were only achieved when latencies were analyzed by the amount of practice that participants had with each individual production (not by practice with the task overall). This finding was consisten t with that obtained by Delaney, Reder, Ritter, and Staszewski (1998), who determined that the power law should be applied to strategies used within a task, not to the task itself as a whole. General problem solving appears, in many ways, to be analogous t o the process of conducting searches within virtual information spaces. In this way, the latter may be seen as simply a special case of the former. The various search heuristics and paradigms employed in problem solving may be identical to those utilized at nodes in information space that require navigational decision making. Depending on the difficulty and linearity of the navigational task at hand, weak methods, such has hill climbing, may be sufficient. For solutions that possess less obvious paths, stronger methods that entail domain specific knowledge may be required. Decisions made in information space, like those in general problem solving, may be a product of upon the current context experienced by the user.

PAGE 19

19 Human Behavior in Information Spaces While immersed in information space, users will experience internal changes. First, they will modify their behaviors based upon whatever objectives they have created o r have been given. Second, they will form mental representations, traditionally known as cognitive maps, of the information space to help guide their navigational behaviors. Goals of Users As computer systems become increasingly larger and more networked the issue of how users can navigate through vast information spaces is critical. Psychologists have focused their space is largely dictated by his or her pr imary objective. That is, what users do depends on the goal of their current task. Benyon and Hook (1997) defined and described the behaviors associated with several of these task objectives. Wayfinding is primarily concerned with reaching a specific destination, or obtaining some specific information. Users, often through exploratory learning, determine a path by which to reach thi s destination. In exploration, users are not trying to reach a specific location. Instead, they are merely interested in scanning and wandering through the information space, exploring the nature and contents of the terrain. During exploration, users ar e primarily concerned with determining which objects are located at which nodes. Finally, in object identification, the user is not concerned with the location of objects or with finding a path to a goal. Instead, the user seeks to identify information a ttributes associated with objects and to find categories and clusters of objects spread across environments.

PAGE 20

20 Cognitive Maps space will occur. Successful navigation lik map, which is a spatial mental representation of the virtual information space. Tolman, Ritchie, & Kalish (1946) argued that rats form cognitive maps during maze learning tasks. Tolman and colleagues set up a very simple maze, whereby one group of rats always found food by making a right hand turn. A second group of rats always found food in one location but whether they made a left or right hand turn was initially uncertain. Using this setup, Tolm an placed behavioral and cognitive views of learning in direct competition with one another. Results learning group) learned much faster group). Therefore, Tolman claimed that what is learned is not behavior but rather knowledge that can be used to direct the behavior. on knowledge of specifi c routes learned through stimulus response reinforcement associations. expectations that certain actions would lead to certain outcomes. For example, rats in maze l earning experiments learn that heading toward a specific location will ultimately lead toward the goal. Neisser (1976) discussed human navigation with regard to cognitive maps that people create of their environment (e.g., a virtual information space). He believed that these maps are not static. Neisser argued that individuals draw important cues from the immediate environment and develop knowledge of the information space over time. These ecological considerations and theories of knowledge evolution hav e formed the basis of several innovative cognitive models of

PAGE 21

21 information processing in HCI and, in particular, information search and retrieval. Such models are described in this paper. Cognitive Models of Search in Familiar Information Spaces A requirem ent of any information processing account of human problem solving and, in particular, of information search, is that it includes a mechanism by which people remember which goals and operators have been evaluated and which have yet to be evaluated, regardl ess of the search task (e.g., Tower of Hanoi, a water jugs problem, a WWW search task). Surprisingly, preeminent architectural accounts of cognition (e.g., Soar, ACT R) do not include such a lity to represent the episodic familiarity of an operator is questionable (Altmann & John, 1999). In ACT R, the base level activation of a chunk represents how recently and frequently it is accessed. However, measures of frequency and recency are combine d into a single representation of activation. That is, the frequency and recency components of base level activation are not independently inspectable by production rules. This severely limits the construction of certain production rules. For example, i t would be impossible to write production rules in ACT R that prefer the most recent operators at the expense of the most frequent operators (Howes & Payne, 2001). The next section summarizes models which do include a clear episodic memory function as a p rimary feature of their architecture. Due to the relatively low number of files stored on PCs (due to hard disk space restrictions) and early limitations in accessing the WWW, much of the early research that has focused on the role of episodic memory in virtual information search has used menu structures. However, menus, like files and directories, are typically organized according to a tree structure. In addition, both structures allow users to navigate in the backwards direction, toward the root node and away from the leaf nodes of the tree. Finally, both structures, unlike huge, vast

PAGE 22

22 information spaces such as the WWW, typically represent fairly small, personal information spaces with which users are likely to become familiar. For instance, users wo uld be expected to commit to memory at least portions of a frequently used, user defined file hierarchy and the Therefore, other than the nuances of how users poi nt and click their way through nodes, and perhaps the semantic wording of labels, there is virtually no difference between searching for a menu item and searching for a file. Card (1984) conducted two experiments in which he explored how users perform visu al search to locate a target command in a menu. In Experiment 1, participants searched for menu items while their eye movements were monitored. Results showed that, while attempting to locate a menu target, participants used an unsystematic search, where by the user may search the same location more than one time. Using these data, Card devised an unsystematic search model, which predicted the distribution of search times, the frequency of saccadic directions, and the lack of an item position effect (i.e. there were no differences among the times to find items at different positions in the menu). In Experiment 2, participants performed visual searches in the same menu. However, in this experiment, menu items were organized according to one of following three arrangements: alphabetic, random, or categorical. Initially, the menu arrangement type influenced the search time required to locate a target command. Specifically, alphabetically arranged menus yielded the fastest search times, followed by catego rically and then randomly arranged menus. However, by the twelfth experimental block, users had generally received enough practice such they had learned the location of each item in the menu, rend ering the arrangement type inconsequential.

PAGE 23

23 ents highlighted the role of episodic memory acquired through practice. However, his study had limited generalizability because of two limitations. First, only half of his participants were experienced mouse users. This may have had a profound effect on search times. Second, the experiment empirically tested items listed serially within a single menu, not a structure of menus. A modern software application almost invariably includes a sophisticated menu structure, organized according to a tree hierarch y. A theoretical limitation of the study is also noteworthy. No mechanism through which menu item locations are committed to memory was provided. It is unclear whether it was spatial episodic memory, verbal (semantic) episodic memory, or a combination t hereof, that facilitated menu search. For example, did users locate fourths of the way down the menu or because they knew that ep isodic memory confine visual search to certain regions of the display and to what extent does episodic memory confine search to certain regions of a file or menu structure? Presumably, the latter mechanism is more foolproof because, depending on display s ettings and viewing options, Richardson, Howes, and Payne (1997) conducted a study that addressed these three shortcomings. They conducted an empirical investigation of episodic memory acquisi tion for routes through menu structures, using a function model called AYN. AYN was developed by Howes (1994) to model the exploratory acquisition of menu knowledge. As it interacts with the menu structure, AYN acquires two types of knowledge: recogniti on knowledge and control knowledge. Recognition knowledge consists of episodic chunks that are encoded into memory for every goal menu action combination that the model encounters, regardless of whether or not the action leads to the goal. Recognition kn owledge is used by AYN to help guide search in the

PAGE 24

24 menu structure both during initial exploration and during subsequent trials. AYN employs the following set of rules to constrain search space: 1) if the goal has not yet been achieved, then avoid recogni zed selections; 2) if the goal has been achieved and there is a recognized selection, then it should be applied; and 3) if there are no recognized selections and the goal has been achieved, then AYN should apply a backup operator in order to locate a recog nized selection. Control knowledge, on the other hand, is acquired through exploration of the menu structure and determines which menu selections lead to the goal and which lead to dead ends. Positive control knowledge refers to movements through the men u structure that lead to the goal. Negative control knowledge refers to movements that lead to dead ends. In AYN, working memory is required to store only the previous action, as opposed to every prior movement made through the menu structure, before the the end of each trial in which the goal is reached, AYN only learns positive control knowledge for the selection that immediately preceded goal attainment. In this way, positive knowl edge is sent back up the structure in a final first manner until positive knowledge has been learned for all selections leading to the goal. Negative control knowledge for selections that lead to dead ends is acquired by AYN in a similar fashion. Richards on and colleagues (1997) reported that computer implementations of AYN can successfully search and learn about menu structures (binary menu trees, in the case of this know ledge accurately modeled two emergent user behaviors: 1) users required fewer and fewer actions to achieve the goal over a number of trials; 2) users remained dependent on the display reliance on recognition behavior results.

PAGE 25

25 Participants were better at backing up from nodes from which they had not made a forward move than they were from backing up from nodes from wh ich they had previously made one or more behavior provided support for the fact that AYN uses recognition knowledge to help determine which option to choose at ea ch selection point. In addition, on trials 3, 4, and 5, participants were more likely to back up from nodes from which they had previously made more than eleven forward moves than from nodes from which they had only made between six and ten forward moves. This acquisition of negative knowledge provided partial support for the fact that AYN acquires control knowledge that determines which actions lead to failure and which actions lead to success. It should be noted, however, that, on all trials, participa nts were much more likely to back up from nodes from which there had been no forward moves at all. Therefore, as is the case with problem solving in general, if any negative knowledge was acquired, then its contribution to performance was much less than p nodes. In a problem space or information space, an individual may re enter a previously visited state either because of a loop or through back up (as described above). Recognition that the state has already been visited and/or that an operator has already been applied to that state helps to prune the search space and constrain the effort spent on attaining the goal (Howes & Payne, 2001). This constraint of search space has been used in a number of models of huma n problem selection process. More familiar operators had a higher p robability of being selected than less familiar operators.

PAGE 26

26 Payne, Richardson, and Howes (2000) investigated the role of familiarity in controlling interactive search by focusing on decision making processes at nodes, or problem states. They found support for the hypothesis that individuals may control search simply by recognizing the actions that have been attempted previously. Specifically, individuals were more likely to select familiar items that led to success on previous trials than unfamiliar items. However, individuals were more likely to select unfamiliar items than familiar items which they knew had previously led to failure (i.e., dead ends). In this sense, the use of familiarity to guide selection was strategic. Using these data, Payne and co lleagues constructed an interactive search model, which, unlike ACT R, relied on the separate and strategic use of information about the frequency function ( and to simulate a lack of reliability on the part of the users), frequency and recency was made to decay stochastically from memory, something that does not occur in ACT R. In addition, false positives were randomly generated in response to queries about whether or not operators had been applied at a particular node on a particular trial. This is consistent with some user behavior, whereby a user may claim to have applied an operator whereas he or she, in actuality, had only seen it and not applied it. Ho wes and Payne (2001) empirically tested their model in a search of simple binary trees. Results exhibited incremental benefit for a search algorithm, like the one employed in their model, that is supported by an episodic memory system and that delivers in dependent estimates of frequency and recency. Results from the simulations correlated significantly with user behaviors. All participants were able to complete the search tasks and three strategies were used: Random search with a forward bias. Participa nts selected either the top or the bottom button on the right of node X, searched the sub tree, and then, on returning to X, selected the other button.

PAGE 27

27 Systematic search. Participants always selected the top button on the first visit to node X, search the sub tree, and then on return to X, selected the bottom button. Memory based search. On trials after the first, participants generally attempted to remember the correct path. On trials after the first, participants flexibly interleaved search based on mem ory for previous trials with, when memory for previous trials failed, either systematic or random search. direct memory retrieval or plausibility judgments in a statement verification task in the sense that a search strategy was superior to a memory strategy in all cases except when episodic memory traces were very strong. None of the participants perseverated. That is, they did not repeatedly search the sa me incorrect sub tree more than several times. With practice (about 4 trials), all participants were able to follow the correct path with relatively few errors. Those participants who used a systematic strategy were significantly more efficient than thos e who did not. On the first trial, the variation in the performance of the systematic participants was less than the variation in the performance of the random participants. All of the models discussed up until this point, whether they address search of l arge unfamiliar information spaces or small familiar information spaces, have dealt with structures that are assumed to be stable over time. That is, the contents of nodes (e.g., their directory names) and the structure which defines how those nodes are c onnected do not change over the period of time during which participants are tested or simulations are run. Factors such as recency and frequency of node visitation can influence internal representations of information space but have no effect on the spac through cost impositions (discussed in the next section) may also serve to change internal, but not external, representations of an information space. However, there are factors, such as size struct ural changes, that alter not only internal representations but also external representations of

PAGE 28

28 the information space. While recency and frequency are the most commonly studied factors, these other factors may also play in important role in determining st rategy during search. The following sections describe how size changes, structural alterations, cost impositions, and other variables may contribute to the adoption of an episodic memory technique versus a search based strategy during target retrieval. Ep isodic Memory V ersus Search During Information Search This section strives to place the current study in context. At any given node within a hierarchically structured information space (e.g., a user defined file structure), an individual must make a decis ion as to which direction to traverse. This decision may be easily and relatively effortlessly achieved via direct retrieval from episodic memory. If the user has previously stored the sought after target, or is otherwise familiar with the information sp ace, then he or she may rely solely on memory of the path taken to reach that target node. On the other hand, if a user is completely unfamiliar with the information space, then he or she may rely exclusively on search parameters (e.g., directory names) w hile performing the target retrieval task. Perhaps the most interesting question concerns the use of both techniques. Under what conditions is the relative contribution of memory greater than that of search, and vice versa? The importance of the recency and frequency of interaction with the information space has already been discussed. As this interaction (with tree nodes, in the case of a file retrieval task) becomes more recent and more frequent, users may rely more heavily on memory than search. In a ddition, a variety of other user characteristics, task characteristics, and characteristics of the virtual environment itself may help to predict the relative contributions of search and memory. With regard to user characteristics, there may be a variety of demographic (e.g., age) and neuropsychological (e.g., spatial ability) variables that play a role. For example, younger adults

PAGE 29

29 and individuals with greater spatial memory ability may rely more heavily on memory than elderly individuals and people with less developed spatial skills, respectively. With regard to task characteristics, users may begin to depend less on memory and to employ more effortful search heuristics as the costs associated with making navigational errors within the information space increase. Furthermore, if the time delay interval between target storage and retrieval becomes sufficiently long, individuals may depend on search attributes to a greater extent, as the target path may no longer be readily accessible via episodic memory. With regard to characteristics of the virtual environment, the size and structure of the information space are likely to be very important. As the size of the information space grows larger and as its structure begins to differ from that with which the i ndividual is familiar, the user may begin to rely more heavily on search techniques than memory retrieval. In many cases, the user may only resort to search techniques when direct retrieval of the target path from episodic memory has failed. Only when t he effects of user, task, and information space characteristics are known can a comprehensive model of target search be formulated. Ultimately, this model should be able to accurately and reliably predict the relative contributions of memory and search he uristics at every node within a given information space during a target retrieval task. There is precedent in the literature for dissociating between two strategies that vary in cost and probability of success. Reder (1982) examined two alternative strate gies that may be used for a forced decision making task. She described a pair of disparate views relating to fact retrieval and plausibility judgment. The traditional view maintains that individuals will attempt to determine the truth of a statement firs t by looking for a close match to the query in memory.

PAGE 30

30 judging its plausibility. The second view, for which Reder argued, holds that plausibility judgment is always a more efficient strategy than direct retrieval, except when verbatim memory is very strong. In her experiment, participants were asked to judge the veracity of various statements relating to stories that they read. Reder gave different groups of participants different reading comprehension questions answerable either by a plausibility judgment strategy or a retrieval strategy. Participants who were initially given a mix of 80% plausibility and 20% retrieval questions were more apt to use plausib ility judgment on subsequent questions. On the other hand, participants who were orig inally given a 20% plausibility 80% retrieval combination were more likely to use the retrieval strategy on later questions. Other researchers have also studied various s trategies, some of which entail calculations and some of which do not, used by participants when faced with a decision making task. Studies the frequency of expos & Ritter, 1992; Schunn, Reder, Nhouyvanisvong, Richards, & Stroffolino, 1997). Schunn and operator, half using an invented operator. The time that participants were given to answer a selection decision (whether or not to compute) was based on problem familiarity rather than on the ability to retrieve some answer, even when problem familiarization occurred 24 hours prior t o testing. The authors argued, therefore, that strategy selection is governed by a familiarity of

PAGE 31

31 retrieving the answer. N that is distinguishing a piece of information as being familiar, may be viewed as a special case of availability. For 30 years, it has been known that individuals use the availability of information to memory as a heuristic for judging the frequency of events and outcomes (Tversky & Kahneman, 1973). However, in certain situations (i.e., when the potential consequences of a decision are grave), one may need to consider more information than is provided by simple recognition or availability of answer. Newell and Shanks presented participants with a series of two alternative forced choice investment decisions between two fictional companies. For each decision, participan ts were provided, at no cost, with the names of the two companies. Three additional pieces of information about each company were available at a cost to the participants. These were the recommendations of three experienced financial advisors (named Richa rd, Tom, and Henry) to invest or not to invest in the companies. Each piece of information was assigned its own validity and discrimination rate. Results showed that, while recognition had a high predictive validity, it was relied on solely by the majori ty of participants. However, when other cues had higher predictive validity, these cues were used and recognition was ignored. These results suggested that recognition is treated merely as a cue among other cues in a decision making task. According to t hese data, recognition should not be attributed a special status, whereby it is viewed as the sole criterion on which the majority of decisions are made. The tasks given by Reder (1982), Newell and Shanks (2004), and Schunn and colleagues (1997) are obviou sly quite different from an HCI task requiring a user to make intermediary traversal direction decisions at each node in order to ultimately locate a target file. However, in

PAGE 32

32 all cases, the participant must make a decision and can do so by using one of tw o alternative strategies, fact retrieval or plausibility judgment (computation). If these perspectives are applied to a file retrieval task domain, one would hypothesize that the user would judge the plausibility of his or her next traversal move via some attribute, such as name of the current node (e.g., name of a current directory). Only if memory of the target path was very strong would the direct retrieval strategy be utilized in deciding upon the next traversal move. By looking at the processes unde rlying the decision made at each node, it would be possible to determine the relative contributions of each of these strategies to retrieving a target file embedded somewhere within a hierarchical structure. s recently and/or frequently been exposed to factual information, then his or her verbatim memory of that information is likely to be very good and direct retrieval will be used when judging the truth of statements pertaining to that information. Similarl y, the recency and frequency with which people visit nodes within a problem, or information, space, are important factors that may dictate future decisions made by theory as a foundation, this paper proposes that memory plays an integral role in information (i.e., success or failure) at various nodes within the tre e structure may strongly influence decisions to be made later at those same nodes. It is further hypothesized that that the relationship between earlier and later decisions generalizes to the relationship between target storage and target retrieval. That is, the path traversed by a user when storing a particular file in a tree structure may activate specific memory traces under certain conditions (e.g., if the file was recently stored and/or the nodes comprising

PAGE 33

33 the target path were previously visited freq uently). If these memory traces remain strongly active, it is proposed that: 1) the user will rely largely, if not exclusively, on episodic memory (as opposed to more effortful search) while searching for the target file, and 2) the user will traverse t he same, or a similar, path when asked to retrieve the target file. This latter proposal implies that the relationship between target storage and retrieval is analogous to context dependent learning observed in transfer appropriate processing (TAP), which posits that test performance is optimal when study and test conditions match (e.g., Lockhart, 2002). Given that only one correct path may be traversed in reaching a destination node, the existence of TAP in a task requiring a user to locate a target file within a tree structure is not readily testable in the laboratory. An experimenter could potentially circumvent this problem by investigating errors committed during traversal of the sole correct path or by embedding the target file in some sort of virtu al maze, instead of a tree, such that the user could reach the target file via multiple paths. Another option would be to augment a tree structure with either shortcuts or hyperlinks in order to provide alternative paths to the same target file. Although this would not alter the underlying file. All of these methodologies would help to determine the relational property between memory trace and the condi tions of path retrieval. deploy a memory based mechanism or a search based mechanism during a file retrieval task. of implementation cost of a problem solving operator, a concept originally introduced within the ACT R framework (Anderson, Matessa, & Lebiere, 1997). This cost may be in the form of time loss incurred (e.g., lockout time), physical and/or mental energy expended (e.g., error recovery cost), and/or task restrictions imposed (e.g.,

PAGE 34

34 keystroke limit). They found that implementation of these types of costs could greatly increase the amount of planning that occurs before a participant begins solving a problem and at each node within the problem, or information, space en route to problem solution. For example, if, following commission of an error, a user must manually enter a command string in order to elicit a particular function, as opposed to the much less e ffortful point and click technique, then she will devote a greater amount of time and cognitive energy to up Payne, 1999). Similarly, participants, for whom a keystroke limit was imposed, exhibited more successful and efficient ex ploratory learning of an information space (i.e., a virtual digital clock) than participants for whom there were no constraints (Trudel & Payne, 1995). Regardless of their type, implementation costs seem to increase planning. In general, more planning le ads to such planning continues until the computed benefits of further planning are outweighed by the costs that would be incurred by continued planning. Associati ng a cost with the implementation of an operator or using a cost to punish a navigational error may be effective ways to study the effects of planning on problem solving in virtual information space. These costs may elicit planning strategies when plannin g might otherwise not occur. For example, users may completely avoid planning when they are first presented with a file retrieval task if they are not familiar with the information space that contains the target. In general, individuals do not plan in no vel and unfamiliar problem solving tasks (Atwood & Polson, 1976; Delaney, Ericsson, & Knowles, 2004). A variety of explanations for this absence of planning have been proposed. Delaney and colleagues, for instance, attribute it to the fact that a full lo ok ahead plan is often not required to solve mundane problems on a

PAGE 35

35 using weak strategies. Weak strategies, such as hill climbing and means ends analysis, in which an individual employs a best guess heuristic as to what next move would bring him or herself closer to the goal state, frequently do not apply to more complex problems (e.g., water jugs problems, target retrieval tasks). In fact, weak methods may not onl y retard but also mislead an water jugs task, employment of a hill climbing heuristic results in irreversible looping moves that actually move the participant f arther from the goal state. Regardless of the reasons behind lack of planning, cost imposition may be used to foster planning and more conservative problem solving strategy (e.g., search in lieu of memory) when it might not otherwise occur. In a m ore recent study of cost consideration, Fu and Gray (2006) found that information seeking costs are often traded off against the utility of information. In short, individuals conserved time at the expense of effort Participants were observed to a dapt to the cost and information structures of environments in a map navigation task and often stabilized at suboptimal levels of performance. A Bayesian satisficing model (BSM) was proposed and implemented in the ACT R architecture to predict information seeking behavior. The BSM used a local decision rule and a global Bayesian learning mechanism to decide when to stop seeking information. The model matched the human data well, suggesting that adaptation to cost and information structures can be achieved by a sim ple local decision rule. The local decision rule, however, often limited exploration of the environment and led to suboptimal performance. The authors concluded that suboptimal performance is an emergent property of the dynamic interactions between cognit ion and the environment. Implementation costs need not be experimentally manipulated. They can also be intrinsically driven. For example, an implementation cost may simply be the opportunity cost of

PAGE 36

36 not having selected a different, more desirable operato r. As already discussed, Lovett and Anderson (1996) argued that problem solvers use the following two sources of information to make operator selections: history of their success (or failure) when applying certain operators and information from the curre nt context of the problem. In their task, for example, BST problem solvers were less likely to apply operators with which they had experienced prior failure and by which the current state would be moved farther from the goal state. A computational model of BST problem solving, developed within the ACT R framework, was devised. It highlighted the manner in which participants weighed the costs associated with selecting various operators at different stages en route to problem solution. Evidence from the L ovett and Anderson study supported the ACT R dissociation between production practice and history of success. Of particular interest were the findings that the absolute number of uses of a production and the proportion of successes with that production af fect different performance measures. In this way, the model of BST problem solving provided a unifying framework in which both types of processes can be integrated to predict the operator selection tendencies of problem solvers. Given th ese findings with regard to costs of operator selection, it can be hypothesized that imposing and then increasing the costs associated with the commission of navigational errors during a file search task would decrease the use of memory based strategies an d increase the use of search based techniques. In order to avoid incurring a cost, a user should tend toward safer, albeit more time consuming, search strategies. This hypothesis could be investigated by manipulating a variety of cost variables, includin g error recovery, lockout time, and keystroke and/or mouse click limits. If an everyday user commits a navigational error during file search, speed disks an d processors. Keystroke and mouse click limits would perhaps be of highest

PAGE 37

37 interest, as they are most germane to the experience of PC end users. A user generally wants to locate a file of interest with a minimal amount of planning and within a minimal nu mber of keystrokes and mouse clicks, both of which serve to minimize the time required to ultimately retrieve the file. Cost imposition would be interesting as it would place desire to minimize planning in direct competition with desire to minimize keystr okes and mouse clicks. Summary and Conclusions With the information era truly upon us, it is more important than ever to understand how human beings organize and retrieve personal data. A deeper comprehension of the human factors associated with informati on search can facilitate the design of more user friendly information browsing tools (e.g., File Explorer, Internet Explorer). This need for understanding is, in many ways, proportional to the advancement of technology. The study of how people store and search for personal files was less important twenty years ago because storage devices simply lacked the capacity to house very large directory trees jam packed with personal data. Similarly, the menu structures of earlier programs were much less intricate and the need for managing bookmarked web pages was undermined by limitations in the size of and access to the Internet. However, as the size and ubiquity of virtual information spaces have increased, so has the need to understand human search and problem solving in those spaces. Traversal of virtual information space (e.g., a hierarchical directory/file tree) in search of a target (e.g., file) may be viewed as a special case of problem solving within the realm of cognitive psychology. For the purpose of the present research, problem space and information space are assumed to be equivalent, at least analogous. Whether information space is referring to items in a user derived menu structure, bookmarks in an internet browser, or files and folders on a perso nal computer, is a relatively trivial matter. An information space may be best

PAGE 38

38 purpose is to allow information to be stored, retrieved, and possibly transf ormed (Benyon & Hook, 1997). A variety of cognitive models and architectures reflect, to varying degrees, the information processing (i.e., search) and episodic (i.e., memory) components of virtual information retrieval. In addition, an assortment of user task, and information space characteristics may determine the relative contributions of episodic memory and search heuristics when searching a hierarchical th interval). Existing models attempt to link episodic memory to the recen cy and frequency factors, particularly when these two factors are measured and considered independently. However, the relationships among episodic memory, search, and the other factors are less clear. Studies that clarify these relationships are required before a comprehensive model of target search can be formulated. Ultimately, this model should be able to accurately and reliably predict the relative contributions of memory and search heuristics at every node within a given information space during a t arget retrieval task. The present study sought to investigate the relative contributions of memory and search to file retrieval in a PC based context. Present Research Rationale discussed above into the digital world. Specifically, I wanted to investigate the cognitive strategies that underlie folder path traversal decisions in a PC based file retrieval task. Without the use of a file exploration tool, file retrieval entails to varying degrees two navigation strategies: memory and search. The memory strategy accesses an episodic memory trace (acquired during prior

PAGE 39

39 interaction with the directory tree structure), indicating the direction or path which the user must traverse in order to move closer to the target location (i.e., file of interest) The search strategy refers to a presumably more effortful process involving reasoned comparisons based on semantic information (i.e., hierarchically organized folder names). In short, navigating toward a file embedded within a user derived, hierarchically organized directory structure entails independent contributions of these two distinct problem solving strategies, one mediated by episodic memory, the other mediated by semantic memory The following experiments sought to develop user profiles unique to the memory and search strategies employed during a file retrieval task. Of particular interest were the conditions for which one strategy dominated over the other. The current investig ator is not aware of any prior research that has explored the relative contributions of search and memory in a file retrieval context. The results of this study have yielded potential implications for the design of software based file exploration tools. Software developers could exploit human tendencies and strengths in the design of file navigation tools by incorporating features well served by such human factors into the user interface of the tools themselves. Experimental Objectives Two experiments we re conducted. Experiment 1 served a normative data collection function in that it selected the verbal stimuli to be used in the subsequent experiment. The objective of this experiment was to populate two binary trees with folder and file names. Specific ally, one tree would be sensible and logical in terms of the folder names that comprise it. The other tree would completely illogical and nonsensical with regard to how the folder and file names are hierarchically organized. Experiment 2 randomly assigned participants to one of four conditions. Two baseline conditions (Memory and Search) established two sets of baseline metrics, one for the memory

PAGE 40

40 strategy and one for the search strategy. That is, experimental manipulations forced participants to use sol ely memory or solely search as their decision making strategy at each node within the hierarchy. The exclusive use of a single exploration strategy generated a user profile specific to that strategy. The end result was two strategy specific user profiles (one for memory, one for search). Two experimental conditions (Memory/Search match and Memory/Search no match) provided a navigation environment in which either or both of the two exploration strategies could be employed in traversal decision processes. By comparing traversal data in these conditions to findings in the baseline conditions, a level by level analysis of strategy type was performed.

PAGE 41

41 CHAPTER 2 MATERIALS AND METHOD S Experiment 1 Participants Forty individuals (twenty five males, fifteen fema les), the majority of whom were students from the University of Florida, at least 18 years of age ( M = 20.44 years, SD = 2.38 years), were recruited for this experiment. Participation was entirely voluntary and withdrawal for any reason was permitted at a ny time. Undergraduate psychology students recruited from the UF participant pool received course credit for their participation, while other participants received no compensation. All participants had normal or corrected to normal vision and were native English speakers. Design and Procedure Each experimental session lasted approximately sixty minutes. Participants rated the semantic relatedness for pairs of potential folder/file names and to gauge the probability that one folder/file name was a subordi nate of another folder name in a personal computer, folder hierarchy context These names were pre selected words or phrases related to the types of information stored in personal file systems ( see Boardman & Sasse, 2004) The rated stimuli included fold were excluded. Participants were instructed to place themselves in the following me ntal context for the name pair a Relatedness score and a Super /Subordinate score. For the Relatedness rating,

PAGE 42

42 how similar the two words or phrases are as folder names on a personal computer. ow likely is it that a folder with one name would contain information that is in general, similar to that which would be contained inside a folder Super/Sub score, unlike the relatedness rating, participants were asked to consider specific sub folders and files ow likely is it tha t the folder [name] on the left would contain the folder/file [name] on the right? H ow likely is it that the folder [name] on the left would be higher in the folder hierarchy on that computer than the folder/file he Likert scales in Table 2 1 to assign the semantic relatedness and super/sub scores to each of the folder/file name pairs. Several examples were provided for clarity. After the data were collected and analyzed, folder names were assigned to two binary t rees, Tree A and Tree B, according to pre established statistical criteria. In short, Tree A was to contain folder name pairs possessing relatively high super/sub scores, while Tree B was to be comprised of folder names with relatively low super/sub score Relatedness score was within a certain range for it to be included in either of the two trees. See Results Experiment 1 for specific inclusion criteria. In more general terms, Tree A made logical, hierarchical sense, w hereas Tree B was hierarchically nonsensical. Figures 2 1 and 2 2 illustrate, by example, the semantic and super/sub relationships within Tree A and Tree B, respectively. These figures merely illustrate portions of the tree; the top node in the figures d oes not represent the root node of the entire tree.

PAGE 43

43 Experiment 2 Participants Forty seven individuals (thirty one females, sixteen males), the majority of whom were students from the University of Florida, at least 18 years of age ( M = 20.15 years, SD = 2.19 years), were recruited for this experiment. Participation was entirely voluntary and withdrawal for any reason was permitted at any time. Undergraduate psychology students recruited from the UF participant pool received course credit for their part icipation, while other participants received no compensation. All participants had normal or corrected to normal vision and were be native English speakers. Design and Procedure Each experimental session lasted approximately 60 minutes. Participants were randomly assigned to one of four conditions: Memory Search Memory/Search match Memory/Search no match The tree structure and size, as well as the navigational options, were identical in all conditions. Both Tree A and Tree B were balanced binary tr ees and contained 7 levels, 128 leaf nodes, and 255 total nodes. The top four levels of this balanced binary tree are depicted in Figure 2 3. For all four conditions and for both experimental phases (storage and retrieval, described below), a participan t was able to traverse in any one of three directions from any given node. Specifically, a user could drill down in one of two forward (i.e., one level deeper) directions (i.e., the binary decision) or choose to drill up in the backward (i.e., one level m ore shallow) direction. Of course, if the participant was currently located at the root folder, a backup moves was not possible; he or she could only navigate in one of the forward (down) directions. Similarly, if the

PAGE 44

44 user was presently residing at one o f the (non target) leaf nodes, he or she was only able to move in the backward (up) direction. No folder/file exploration tool or any other means by which the participant could gain a r retrieval phases for any of the conditions. At any given node, participants were only able to see the node at which they currently resided, although they were able to preview all three proximal nodes (i.e., the two forward nodes and the one backward nod e). Figures 2 4 and 2 5 show the interactive displays used by participants during Experiment 2 sessions. The screen shots were taken from the practice mode. However, the practice mode differed from the experiment mode only in terms of the types of folde r names used (animals versus personal information management files) and the number of to be stored/retrieved files (4 versus 16). Any given condition was different from all other conditions in one or both of the following ways: 1) the presence or absence of a storage phase, and 2) the presence (or absence) of a semantically related, super /subordinate organization of directory names within the binary tree (Tree A or Tree B) by which users could (or could not) employ reasoned comparisons while navigating t he tree. A summary of the how the four file retrieval conditions can be distinguished on these two factors can be found in Table 2 2. Of the four conditions, the first two conditions (Memory and Search) represented baseline conditions, while the last two conditions (Memory/Search match and Memory/Search no match) referred to the experimental conditions of interest. Participants in the Memory condition interacted with Tree B, while participants in the three other conditions interacted with Tree A. Partic ipants stored files in the Memory, Memory/Search match, and Memory/Search no match conditions, but not in the Search condition. For those conditions in which there was a storage

PAGE 45

45 phase, users stored sixteen files. The order in which these files were store d and retrieved was randomized. The following is a more in depth description of each condition. Memory condition (baseline) Participants in this condition were constrained to using exclusively memory as they navigated toward target files. After completin g a practice session to familiarize themselves with the interface and navigation controls, users were presented and asked to store sixteen files, one by one, in random order. Beginning at the root node (i.e., Level 0 folder), participants stored the files in any leaf node (i.e., Level 7 folder) of their choice within the binary tree containing directory names that were not logically organized according to a super /subordinate node structure (Tree B). In this way, participants were forced to rely exclusive ly on episodic memory traces acquired during file storage when interacting with the tree. 7 folder of choice. P articipants were given a maximum of one minute or thirty mouse clicks (whichever came first) to store each file and were informed of these limits. If file storage exceeded either of these two restrictions, the next to be stored file was administered. Use After all sixteen files were stored (or were attempted to be stored), participants were given an operation span (OSPAN) test. The OSPAN is a quick, general, and wide ly used measure of one at a time, while solving simple math problems. The test contained twelve sets of trials, each containing between two and five words to be m emorized. The OSPAN was such scored such that participants were awarded the same number of points as there were items in a perfectly recited set. So, a perfectly recited set comprised of five words was awarded five points. A

PAGE 46

46 perfect recitation entailed not only recalling every word in a set but also recalling every word in OSPAN score. Administration of the OSPAN not only provided a measure of individual working memory but also served as a time delay between storage and retrieval, thus enabling a true comparison of memory and search strategies with regard to file retrieval efficiency. Following completion of the OSPAN, participants were asked to retriev e the same sixteen files they had originally stored in the same order in which they were originally administered. Participants once again began at the root folder and progressed deeper into the tree toward the leaf nodes in search of the target file. Giv en the nature of a binary tree, there was only one correct traversal path between the Level 0 root node and the Level 7 target file. Unlike storage, retrieval of a given file was accomplished by navigating to the Level 7 folder location at which it was st ored. No dragging and dropping or other mouse controls were required for file retrieval. Users were again given one minute or thirty mouse clicks (whichever came first) to successfully retrieve each target file and were informed of these restrictions. I f the target file had not been retrieved within the allotted time or number of mouse clicks, the next to be retrieved target file was administered. Search condition (baseline) This condition was methodologically very similar to the Memory condition. The only two differences were that the Search condition did not consist of a storage phase and that users interacted with Tree A, the logical tree, instead of Tree B. In this way, participants were constrained to using, at least initially, only search when i nteracting with the tree. The absence of a storage phase prevented the formation of episodic memories related to correct traversal paths within the tree prior to retrieval. In this way, participants were forced, at least initially, to rely

PAGE 47

47 solely on reas oned comparisons based on hierarchically organized semantic information when making their navigation decisions. Memory/Search match condition (experimental) This condition was procedurally identical to the Memory condition. The only difference was that Memory/Search match participants interacted with Tree A, the same logical tree experienced by Search users. In this hybrid condition, participants had the opportunity to use either or both cognitive strategies for their navigation decisions instead of jus t a single exploration strategy. That is, unlike the previous two conditions, users were permitted to use either memory or search, or any combination thereof, while retrieving target files. In this condition, stored files matched to be retrieved files, j ust as in the Memory condition. For the Memory/Search conditions, a brief post task interview was conducted. Participants were asked to estimate the percentage of their overall retrieval strategy that was based on up question, participants were asked, in the cases where memory was used, what percentage of that memory strategy was based on symbolic memory). Memory/Search no match condition (experimental) This condition was exactly the s ame as the Memory/Search match condition. The only difference was that, in the Memory/Search no match condition, stored files did not match to be retrieved files. The sixteen to be retrieved files in this condition were the same as those in the other thr ee conditions (and were the same in both trees). However, in the Memory/Search no match condition, the stored files (and file names) were different than the to be retrieved files

PAGE 48

48 (and file names). In an attempt to keep acquisition of tree familiarity con sistent across conditions, the anticipated storage locations for each of the to be stored files in the Memory/Search no match condition were, distance wise, very near to the to be retrieved file. In fact, they always shared the same Level 6 parent folder, meaning that their horizontal offset was always equal to 1. The purpose of the Memory/Search no match condition was to control for the notion that participants may gain two qualitatively different types of experience (i.e., memory) of the tree one rel ated to a specific path traversed to reach a specific target file, another related to a general familiarity/experience with the tree. This condition was included in order to rid the experiment of potential contamination produced by general experience acqu ired while interacting with the information structure. Summary of file retrieval conditions Table 2 2 summarizes the four file retrieval conditions in Experiment 2. As seen in Table 2 2, three of the four conditions entailed file storage. For the two M emory/Search hybrid conditions (but not for the Memory condition), each file had a leaf node at which it was intended to be stored. However, participants stored some files in unintended locations. These storage pped from analyses in an attempt to minimize data loss and maximize external validity. If a user committed a file storage error on a particular storage trial, the file remained where the user stored it for the remainder of the storage phase and the entire ty of the retrieval phase. The intended leaf node was simply left blank (i.e., not assigned a file name) and participants were not permitted to navigate to it. Storage errors generated differences among participants with regard to node visitation frequen cies prior to the retrieval phase. Statistically, specific correlations were calculated to determine the relationship, if any,

PAGE 49

49 between node visitation frequencies in the storage phase and the speed and accuracy of traversal decisions in the retrieval phas e. These data are presented in the Results and Discussion. Binary Tree Specifications A list of the folder and file names used in Trees A and B can be found in Appendix A and Appendix B, respectively. For identification purposes, each node in the tree can be addressed using a two dimensional coordinate system. The first digit in the coordinate indicates the level within the tree, while the second digit represents the position within that level. For example, node (0, 1) would refer to the first (and o nly) node in Level 0 (i.e., the root node of the tree). Node (5, 20) would indicate the twentieth node in Level 5. A more comprehensive depiction of the node numbering system is illustrated in Figure 2 6. In an attempt to keep tree familiarity acquisi tion consistent across participants, the intended storage and retrieval locations for the sixteen target files were evenly distributed across the leaf nodes of Level 7. That is, four target files were intended for storage/retrieval in each quartile (thirt y Since participants began at the root node for each file trial, any given participant became disproportionately more familiar with shallower tree levels than with deeper levels throughout the storage or retr ieval phase. For example, as shown in Table 2 3, if a participant demonstrated perfect accuracy (i.e., did not execute a single backup move) for the entire storage (or retrieval) and 4, 50% of The ideally performing participant would experience the root node sixteen times (since there were sixteen to be stored/retrieved files), each Level 1 node eight t imes, each Level 2 node four times, each Level 3 node two times, and each Level 4 node one time, as all of those nodes constitute correct traversal paths. However, the participant would only experience each correct

PAGE 50

50 traversal node a single time for Levels 5 through 7 and, as seen in Table 2 3, not all of those nodes comprised correct traversal paths. Level 7 was the only level the tree that consisted exclusively of files. All other levels consisted exclusively of folders. Table 2 4 lists the file name s, descriptions, and locations for all Level 7 stored and retrieved target files for all four conditions.

PAGE 51

51 Table 2 1. Semantic relatedness and super/sub rating scales for Experiment 1 Relatedness rating scale Super/sub scale 0: Not at all related 0: Not all likely 1: Somewhat related 1: Somewhat likely 2: Moderately related 2: Moderately likely 3: Very related 3: Very likely Figure 2 1. Portion of Tree A. Figure 2 2. Portion of Tree B.

PAGE 52

52 Figure 2 3. Top four levels of balance d binary tree Figure 2 4. Storage phase user interface with navigation controls (practice mode).

PAGE 53

53 Figure 2 5. Retrieval phase user interface with navigation controls (practice). Table 2 2. Summary of file retrieval conditions for Experime nt 2. Condition Tree File storage? Memory (Baseline) B (nonsensical) Yes Search (Baseline) A (logical) No Memory/Search match (Experimental) A (logical) Yes Memory/Search no match (Experimental) A (logical) Yes

PAGE 54

54 Figure 2 6. Binary tree coordinate system used for node identification Table 2 3. Stored and retrieved file names, descriptions, and locations for all four conditions. Level Number nodes in level along correct traversal paths Total number of nodes in level Percentage of level experience d Number visitations for each correct traversal node in level 0 1 1 100% 16 1 2 2 100% 8 2 4 4 100% 4 3 8 8 100% 2 4 16 16 100% 1 5 16 32 50% 1 6 16 64 25% 1 7 16 128 12.5% 1

PAGE 55

55 Table 2 4. Stored and retrieved file names, descriptions, and loca tions for all four conditions. Stored files in the Memory condition and the Memory/Search match condition (file name, description, and location) Stored files in the Memory/Search no match condition (file name, description, and location) Retrieved files in all four conditions (file name, description, and location) 1 This file is a copy of your January 2007 cell phone bill. This file is a copy of your December 2006 cell phone bill This file is a c opy of your January 2007 cell phone bill. File location: (7, 6) File location: (7, 5) File location: (7, 6) 2 Statement.pdf Payment History.doc Statement.pdf This file is a copy of your MasterCard September 2006 stateme nt. This file lists all of your payments of previous MasterCard balances. This file is a copy of your MasterCard September 2006 statement. File location: (7, 12) File location: (7, 11) File location: (7, 12) 3 Software Costs.doc Packet Costs.doc Sof tware Costs.doc This file is a list of the money you spent on software that accompanies some course packets. This file is a list of the money you spent on course packets. This file is a list of the money you spent on software that accompanies some cours e packets. File location: (7, 24) File location: (7, 23) File location: (7, 24) 4 Payment Receipt.jpg Citation Total.doc Payment Receipt.jpg This file is a copy of the payment receipt for your on campus parking pass. This file contains the amount o f money that you owe for on campus parking citations. This file is a copy of the payment receipt for your on campus parking pass. File location: (7, 26) File location: (7, 25) File location: (7, 26) 5 pg This file is a picture of you and your friends in Key West during Spring Break 2005. This file is a picture of you and your friends in the Bahamas during Spring Break 2004. This file is a picture of you and your friends in Key West during Spring Break 2005. File location: (7, 39) File location: (7, 40) File location: (7, 39) 6 Shakira Lie.mp3 Britney Spears Everytime Lyrics.html Shakira Lie.mp3 This file is an MP3 audio file. This file is an MP3 audio file. This fil e is an MP3 audio file. File location: (7, 41) File location: (7, 42) File location: (7, 41)

PAGE 56

56 Table 2 4. Continued Stored files in the Memory condition and the Memory/Search match condition (file name, description, and location) Stored file s in the Memory/Search no match condition (file name, description, and location) Retrieved files in all four conditions (file name, description, and location) 7 Freshman Year Entries.txt Sophomore Year Entries.txt Freshman Year Entries.txt This file co ntains diary entries from your freshman year of high school. This file contains diary entries from your sophomore year of high school. This file contains diary entries from your freshman year of high school. File location: (7, 53) File location: (7, 5 4) File location: (7, 53) 8 James Joyce Clay.pdf Nikolai Gogol The Cloak.pdf James Joyce Clay.pdf This file is a short story for recreational reading. This file is a short story for recreational reading. This file is a short story for recreation al reading. File location: (7, 59) File location: (7, 60) File location: (7, 59) 9 References.Doc Abstract.doc References.Doc This file contains the references for your Behavioral Analysis term paper. This file contains the abstract (introduction) for your Behavioral Analysis term paper. This file contains the references for your Behavioral Analysis term paper. File location: (7, 70) File location: (7, 69) File location: (7, 70) 10 Man: The Social Animal.pdf Skeletal Biological Distances.pdf M an: The Social Animal.pdf This file is an article that was assigned in one of your classes. This file is an article that was assigned in one of your classes. This file is an article that was assigned in one of your classes. File location: (7, 76) File location: (7, 75) File location: (7, 76) 11 Psi Chi Requirements.html Pi Gammu Mu Regional Members.xls Psi Chi Requirements.html This file lists the requirements to become a member of this psychology honors society. This file lists the members of thi s social science honors society that you are in. This file lists the requirements to become a member of this psychology honors society. File location: (7, 88) File location: (7, 87) File location: (7, 88) 12 Rehearsal Schedule.doc Members List.doc R ehearsal Schedule.doc This file contains the rehearsal schedule for the Dance Club that you are in. This file lists the members of the Dance Club that you are in. This file contains the rehearsal schedule for the Dance Club that you are in. File locat ion: (7, 90) File location: (7, 89) File location: (7, 90)

PAGE 57

57 Table 2 4. Continued Stored files in the Memory condition and the Memory/Search match condition (file name, description, and location) Stored files in the Memory/Search no match condition (file name, description, and location) Retrieved files in all four conditions (file name, description, and location) 13 Confirmation #.html Test Location.html Confirmation #.html This file contains the confirmation number provided to you when you regist ered for the GRE General. This file contains the test location provided to you when you registered for the GRE General. This file contains the confirmation number provided to you when you registered for the GRE General. File location: (7, 103) Fi le location: (7, 104) File location: (7, 103) 14 This file is a letter written by Dr. White recommending you for graduate school. This file is a letter written by Dr. Fischler recomme nding you for graduate school. This file is a letter written by Dr. White recommending you for graduate school. File location: (7, 105) File location: (7, 106) File location: (7, 105) 15 FL License Requirements.html Real Estate Courses List.html FL License Requirements.html This file lists the state license requirements for becoming a real estate agent in Florida. This file lists the course requirements for becoming a real estate agent in Florida. This file lists the state license requirements for becoming a real estate agent in Florida. File location: (7, 117) File location: (7, 118) File location: (7, 117) 16 Receptionist Intern Posting.html Psychometrician Intern Posting.html Receptionist Intern Posting.html This file is a posting for a paid internship as a receptionist in a psychologist's private practice. This file is a posting for a paid internship as a test administrator in a psychologist's private practice. This file is a posting for a paid internship as a receptionist in a psycholo gist's private practice. File location: (7, 123) File location: (7, 124) File location: (7, 123)

PAGE 58

58 CHAPTER 3 RESULTS AND DISCUSSION Experiment 1 Experiment 1 was conducted to select folder names to populate the two binary trees used in Experiment 2. As a result of this norming experiment, nodes comprising at least the correct traversal paths of both Trees A and B were populated with folder names. All other folder names were hand selected by the experimenter. Tree A was to include only folder names that were, pair wise, related in a super /subordinate structure, while Tree B was to include only folder names that were, pair wise, not related, in a super /subordinate structure. Both trees were to possess comparable degrees of semantic relatedness amon g their constituent folder names. A total of 250 pre selected, potential folder name pairs were rated by participants. The distributions of the Relatedness and Super/Sub scores were graphed and visually inspected. Both sets of assigned scores were determ Relatedness distribution exhibited a skewness and kurtosis of 1.84 and 4.84, respectively, .62, respectively. Tree 2.19 and and 1.36, respectively. Therefore, z score cutoffs were used as criteria when decidin g which folder names to include as Experiment 2 stimuli and which to exclude. The mean Relatedness and Super/Sub scores for each of the two trees, as well as their inclusion criteria, are summarized in Table 3 1. Standard deviation values are provided in parentheses. A list of the folder names that met both sets of criteria and that were selected for inclusion can be found for Trees A and B in Appendix A and Appendix B, respectively.

PAGE 59

59 Experiment 2 The two main dependent measures in the study were time on node (TON), which measures speed, and the probability of executing a backup move (P(BACK)), which measures accuracy, or more precisely, the likelihood of committing an errant drill down move. A check for skewness and kurtosis revealed that TON data were positively skewed, both overall and when separated by condition, as is often the case with reaction time data (e.g., Delaney et al. 1998). Therefore, the TON data were log transformed in order to approximately normalize the distributions, satisfy the und erlying assumptions of numerous inferential statistical tests, and gain statistical power to detect significant differences among conditions and tree levels. The log transformation reduced the overall skewness and kurtosis of the TON distribution from 4.5 6 to .59 and from 39.38 to .72. At each node, only two forward moves were possible. Therefore, once people made an error at a given node, they could quickly choose the other option without having to rely on memory of the storage phase or search. Becaus e the current project was focused on understanding memory and search times, subsequent analyses were restricted to the first visit to a node unless otherwise noted. This was done so as not to artificially decrease TONs by averaging in fast reaction times that were not based on memory or search. Participants became somewhat familiarized with the tree during the storage phase of the experiment (at least in those conditions for which there was a storage phase). It was suspected that tree familiarity gained d uring storage might affect TON in the retrieval phase. If so, a separate analysis involving frequency of node visitations during storage would have to be conducted. However, tree familiarity acquired during the storage phase affected neither TON nor P(BA CK) in the retrieval phase. A near zero correlation was obtained between node visitation frequency during storage and the log transformed TONs associated with first visits to those nodes in the retrieval phase ( r = .05, ns ). Similarly, storage visitation frequency and the

PAGE 60

60 probability of backup moves at the first visitation to those nodes during retrieval were uncorrelated ( r = .06, ns ). Therefore, storage visitations were not regressed out of analyses involving retrieval phase first visitations. Overall Findings by Condition: Speed and Accuracy of Traversal Decisions The first analysis explored whether the four conditions differed overall in the amount of time required to make traversal decisions, which ultimately translates into the total time needed t o retrieve individual files. In statistical terms, this analysis sought to investigate a potential main effect of Condition. Although analyzing speed and accuracy globally by condition was not a primary objective of the current research, presenting these data helps to situate the later results in a condition centric fashion. It is important to remember that a full understanding of how participants performed on the file retrieval task requires taking level of the tree into consideration (see the Results s these analyses.) The objectives of the analyses in the present section were: 1) to identify any overall differences between search performance and memory performance, and 2) to determine whether traversal decisions made in the experimental conditions, where both memory and search strategies were available, were overall faster, slower, or equally quick to the decisions made in the baseline conditions, where only one or the oth er strategy was available. I initially expected that TON in the experimental groups would fall somewhere between that of the Memory and Search conditions, reflecting a mixture of the two strategies. However, while a one way between subjects ANOVA of log TON revealed highly significant differences between the four conditions, F (3,3053) = 97.32, MSE = 5.76, 2 = .09, p < .01, the initial predictions were not supported. Figure 3 2 gives the mean log TON for each of the four conditions. Tukey HSD post hoc tests showed that each experimental memory/search hybrid condition had faster TONs than both baseline conditions ( p < .01) (see Figure 3 1), indicating

PAGE 61

61 that traversal decisions were facilitated when both cognitive strategies were available as opposed to wh en only one or the other was available. There was no difference between the two baseline conditions, suggesting that memory was no faster than search as a decision making strategy, and vice versa. However, decisions in the Memory/S earch match condition were made more quickly than those in Memory/S earch no match ( p < .01). This suggested that familiarity of the to be retrieved file name aided participants, since the retrieved files matched the stored files in the former but not in the latter. The next analysis explored how accurate people were at retrieving files. Again, this was initially done overall by condition to provide a condition based framework for the interaction tion by Level identify any overall differences between search performance and memory performance, and 2) to determine whether traversal decisions made in the experimental conditions, where both memory and search strategies were available, were overall more, less, or equally accurate to the decisions made in the baseline conditions, where only one or the other strategy was available. Originally, I hypothesiz ed that the accuracy of the experimental groups would fall somewhere between that of the Memory and Search conditions, again reflecting a mixture of the two strategies. However, while a one way between subjects ANOVA of log TON revealed highly significant differences between the four conditions, F (3,3632) = 281.18, MSE = 25.18, 2 = .19, p < .01, the initial predictions were not fully supported. Figure 3 3 depicts the mean P(BACK) for each of the four conditions. Tukey HSD post hoc tests revealed that backup moves were much more likely to be executed in the M emory baseline cond ition than any of the other three conditions (all p s < .01), suggesting that as a strategy, memory is far more prone to

PAGE 62

62 error than search. In addition, Memory/Search match was marginally more accurate than Search ( p = .08), and t he Memory/S earch match con dition was significantly more accurate than t he Memory/S earch no match condition ( p < .05), again indicating that file name familiarity played a role in facilitating traversal decisions. These results are illustrated in Figure 3 2. Figures 3 1 and 3 2 s uggest that p articipants were using a combination of the memory and search strategie s available to them in the two Memory/S earc h conditions. I originally proposed to try to determine whether people were relying on search or memory for nodes within the two experimental conditions (i.e., Memory/Search match and Memory/Search no match). However, the assignment of memory/search values was impo ssible because both experimental conditions yielded faster and more accurate decisions than either baseline condition did. Therefore, accurately assigning a memory/search value to each node within the tree was not feasible Consequently, subsequent results will be presented in a condition centric fashion, focusing on differences among st the Memory and Search baseline co nditions and the experimental Memory/Search hybrid conditions. Overall, search produced decisions that were more accurate but no faster than memory. Because it enhanced accuracy without degrading speed relative to memory, search appears to be the better of the two strategies at first glance. Overconfidence in memory might account for this partial speed accuracy tradeoff and overall memory inferiority. Participants may have unjustifiably experienced a feeling of remembering the correct traversal path whe n in fact they had not, resulting in errant moves. Overall Memory and Search speeds may have been comparable for a number of reasons. Although possible explanations include cognitive processes that consumed approximately equal amounts of time and the fac t that participants were informed of click and time limitations in both conditions, the equivalence may be spurious. That

PAGE 63

63 Search speeds. Therefore, before d rawing final conclusions regarding the relative speed and accuracy of memory and search, it is highly recommended that the reader take into consideration the n ext three sections of the Results. As mentioned, true insight into file retrieval task performance emerges only once the and 6) ar e discussed in this section. Nodes comprising these levels do not involve simple binary or back decisions based on solely on memory and/or search. Therefore, these idiosyncratic characteristics should be taken into consideration when reading and interpre ting all results involving those tree levels. Planning time (level 0) Level 0 contains one node, the root node, or parent directory. Only two possible moves were available during Level 0 decisions because no backup move could be made from the root node The Level 0 TONs were practically identical for the Memory and Memory/Search match conditions (mean TON: ~.59 seconds). Similarly, the Level 0 TONS were virtually the same for the Search and Memory/Search no match conditions (mean TON: ~.86 seconds). Post hoc tests p < .01). Regardless of condition, the root node represented a planning phase before drill down began, as reflected in the longer Level 0 TONs r elative to the TONs at all other levels ( p < .01). It was as if participants pre planned or partially pre planned a traversal route when they were presented with a to be retrieved file before proceeding to drill down. However, this planning

PAGE 64

64 period was di sproportionately longer for conditions in which to be retrieved file names were unfamiliar (i.e., search baseline and memory/search no match ). In t he search baseline condition to be retrieved files were unfamiliar because there was no storage phase. In t he memory/search no match condition to be retrieved files were unfamiliar because the to be retrieved files did not match stored files. Less planning seemed necessary for conditions in which to be retrieved files were familiar, that is, where to be retr ieved files matched stored files (i.e., memory baseline and memory/search match ). Leaf nodes (Level 7) Level 7 nodes consisted only of files, including the target files. This was the only level the tree that was comprised of files instead of folders. No forward moves were possible from Level 7; only backup moves could be executed from the leaf nodes. This was the only level of the tree that was comprised of files instead of folders. Level 7 TONs were also comprised of factors other than decision making processes. At Level 7, all retrieval data associated with decisions leading to the retrieval of a target file were purged, as these TONs would merely indicate how e following to be retrieved target file. However, even with these purged data, Level 7 decisions refer exclusively to backup moves that were executed, usually quite quickly, in response to drill down traversal errors. Level 6 The analyses in the section s that follow focus exclusively on the Log TON and P(BACK) dependent variables, respectively, and primarily on Levels 1 through 6 of the tree. However, it should be noted that, while included in the analyses, Level 6 data was somewhat complicated by the f act that participants had an additional strategy available to them visual search afforded by the forward preview nodes. Therefore, at Level 7 in t he memory/search match condition for

PAGE 65

65 example, participants could use visual search, in addition to the memor y and search (i.e., folder name logic) strategies. In this condition, users had a total of three strategies available to them at Level 6, whereas they only two available at every other level. Because Level 7 is excluded from most analyses, Level 6 is gene rally considered the deepest level of the tree. Therefore, when making shallow versus deep comparisons, Level 1 is often compared to Level 6. In lieu of using regression line slopes, ANOVA post hoc tests were used to investigate shallow deep differences. Both regression lines and post hoc tests would have provided information regarding the magnitude and direction of speed and accuracy changes that occurred as users drilled deeper into the tree. However, what post hoc tests provide that regression slopes do not is the ability to detect statistically significant performance differences among levels and among conditions at a given level. Speed of Traversal Decisions: Condition by Level Interaction This analysis was conducted to determine how traversal de cision time as a function of tree level differed among the four conditions. In statistical terms, this analysis sought to investigate a potential Condition x Level interaction. (Recall that results pertaining to the main effect of Condition were reported a priori theoretical question and was therefore not analyzed.) The specific objectives of this analy sis were: 1) to compare how decision speeds changed, if at all, as users drilled into deeper tree levels for the Memory and Search baseline conditions, and 2) to compare the experimental hybrid conditions, not only to each other, but to the baseline perfo rmances. In this way, we could qualitatively liken the speed performances of the Memory/Search conditions to the baseline speed performances using the general morphologies of each of the condition based curves

PAGE 66

66 Figure 3 3 shows TON separated by condition and level of the tree. It was initially hypothesized that, even after tree level was accounted for, the speed of the experimental groups would fall somewhere between that of the Memory and Search conditions, reflecting a mixture of the two strategies. Ho wever, while a two way mixed factorial ANOVA of log TON revealed a significant Condition x Level interaction, F (21,3053) = 4.247, MSE = .23, 2 = .03, p < .01, the initial predictions were not supported. The degree to which the four conditions differed from one another with regard to node to node traversal speeds depended on the specific level of the tree at which a particular decision was mad e, but the interaction was not as expected. I will follow up and discuss the findings in Figure 3 3 first with respect to the baseline conditions followed by the experimental conditions. Tukey HSD post hoc tests were used to identify significant contrast s in each case. Baseline findings With the exception of Levels 0 and 7, t he M emory baseline condition performance, as measured by TON, was nearly perfectly flat. No significant differences existed between or among Levels 1 and 6 for that particular condit ion. This suggested that when memory was employed as a file retrieval strategy, TONs were insensitive to current level of depth within the tree. The M emory baseline condition was faster than t he S earch baseline condition but only at more shallow tree lev els (Level 1: p = .07; Level 2: p < .05). Performance differences between these conditions were based primarily on the different shapes of the Memory and Search curves. Unlike the M emory baseline condition the S earch baseline condition exhibited TONs tha t steadily decreased as users drilled deeper into the tree. A Level 1 Level 6 comparison (Tukey HSD post hoc test) revealed a significant mean TON difference of .33 seconds ( p < .01). This finding indicates that, unlike memory, search exhibited a distanc e to goal sensitivity and enjoyed a steady speedup as participants traversed into deeper levels of the tree. A funnel effect,

PAGE 67

67 whereby users were able to increasingly prune search space as folder names became more specific to the target files, may have bee n responsible for this speedup. Experimental findings Results showed that performance in t he Memory/Se arch match and Memory/Search no match condition s, as measured by TON, was quite flat across Levels 1 through 7. In fact, no statistically significant di fferences existed between or among any of those levels for either of those two conditions. This suggested that having both memory and search available resulted in a performance that, at least qualitatively from a TON perspective, behaved very similarly to Memory. Although the availability of episodic memory traces sped up performance relative to baseline, hybrid condition decisions did not accelerate as users delved into deeper tree levels. Given the morphological differences in the Memory and Search cur ves, hybrid condition Memory/S earch conditions exemplified the complex and non additive presence of the baseline conditions in the hybrid conditions. Together, these results i mplicate a flexible choice or race type model between memory and search when both strategies are available to a PC user during file retrieval. These potential models are discussed in the General Discussion Although not statistically significant, there was a general trend for t he Memory/S earch no match condition to outperform t he Memory/S earch match condition at more superficial tree levels (i.e., Levels 1 and 2). However, the trend crossed over at deeper levels; t he Memory/S earch match condition outper formed t he Memory/S earch no match condition at Levels 3 through 7. This Memory/S earch match condition advantage grew larger at levels more proximal to the leaf nodes and the match no match differences even became statistically significant at Levels 4 thro ugh 6 ( p < .01). In the trees used in this experiment, like most hierarchical trees, folder names are more general at more shallow regions and become

PAGE 68

68 progressively more specific at deeper levels of the tree. These findings suggest that generally termed f older names higher in the tree did not give preferential advantage to the match condition over the no match condition. That is, decision times were relatively fast regardless of whether or not retrieved file names matched the stored file names. Only at d eeper tree levels where folder names were more specific did the match condition facilitate performance relative to the no match condition. Accuracy of Traversal Decisions: Condition by Level Interaction This analysis was conducted to determine how travers al accuracy as a function of tree level differed among the four conditions. In statistical terms, this analysis sought to investigate a potential Condition x Level interaction. (Again, recall that results pertaining to the main effect of Condition were r a priori theoretical question and was therefore not analyzed.) The specific objectives of thi s analysis were: 1) to compare how decision accuracies changed, if at all, as users drilled into deeper tree levels for the Memory and Search baseline conditions, and 2) to compare the experimental hybrid conditions, not only to each other, but to the bas eline performances. In this way, we could qualitatively liken the accuracy performances of the Memory/Search conditions to the baseline accuracy performances using the general morphologies of each of the condition based curves (see Figure 3 4). It was ini tially hypothesized that, even after tree level was accounted for, the accuracy of the experimental groups would fall somewhere between that of the Memory and Search conditions, reflecting a mixture of the two strategies. However, while a two way mixed fa ctorial ANOVA of P(BACK) revealed a significant Condition x Level interaction, F (21, 3632) = 6.721, MSE = .55, 2 = .04, p < .01, the initial predictions were not supported. In deed, the degree to

PAGE 69

69 which the four conditions differed from one another with regard to node to node traversal accuracies depended on the specific level of the tree at which a particular deci sion was made. However, the interaction, depicted in Figure 3 4, was not as expected. As with TON, the P(BACK) data will be discussed separately for the baseline conditions and then the experimental conditions. Tukey HSD post hoc tests were again used t o identify significant contrasts. Baseline findings There was a general trend for the error committal rate in t he M emory baseline condition to increase as users progressed from shallow to deeper tree levels. In fact, a Tukey HSD post hoc test revealed a s tatistically significant Level 5 Level 6 P(BACK) difference. A pair wise comparison of shallow and deep levels revealed a large disparity between the Level 1 P(BACK) mean (.21) and the Level 6 P(BACK) mean (.72) ( p < .01). Furthermore, at all levels deep er than Level 1, t he M emory baseline condition exhibited P(BACK) rates higher than those in t he S earch baseline condition ( p < .01). This once again illustrates the error proneness of memory relative to search. This memory specific error proclivity seems to worsen at levels more distal from the root node, a finding which may be conceptualized in terms of the primacy and recency effects observed in many memory recall studies. Specifically, in a file retrieval context, the made at nodes proximal to the root node) may have a much leaf node), which actually appears to be altogether absent. Again, this possible explanation is prov with studies on interference and serial position effects. Glanzer and Cunitz (1965), for example, distraction between study and test completely eliminated the re cency effect. In the present task, post storage interference may have been comprised of the OSPAN test and the storage and retrieval of other files.

PAGE 70

70 Unlike t he M emory baseline condition S earch baseline accuracy, as measured by the probability of executin g a backup move, was nearly perfectly flat across Levels 1 through 6. No significant differences existed between or among those levels for that condition. This suggested that, unlike memory driven decisions, the probability of committing an error using s earch as a file retrieval strategy was insensitive to current level of depth within the tree. Experimental findings Results showed that performance in the Memory/S earch conditions, as measured by P(BACK), was quite flat across Levels 1 through 6. In fact with the exception of Level 6 in t he Memory/S earch match condition no statistically significant differences existed between or among any of those levels. This suggested that having both memory and search available resulted in a performance that, at lea st qualitatively from a P(BACK) perspective, behaved very similarly to search. Although the availability of search (i.e., folder name logic and reasoning heuristics) sped up performance relative to baseline, hybrid condition decisions did not accelerate a s users delved into deeper tree levels. Given the morphological differences in the Memory and Search curves, hybrid earch conditions exemplifie d the complex and non additive presence of the baseline conditions in the hybrid conditions. Accuracy wise, the hybrid conditions consistently outperformed Search and Memory alone, particularly at more superficial tree levels (i.e., Levels 1 through 3) ( p < .05). Although not statistically significant until Level 6 ( p < .05), there was a trend for t he Memory/S earch no match condition to outperform t he Memory/S earch match condition accuracy wise. This trend became more pronounced at deeper tree levels. T his finding may be attributed to the notion that participants became more cautious and less careless in their decisions, especially at nodes more proximal to the target file, when they realized that the to be

PAGE 71

71 retrieved files did not match the store files. That is, perhaps the file name mismatch in t he Memory/S earch no match condition triggered additional cognitive processing and fostered greater decision making inhibition (i.e., greater cautiousness) relative to t he Memory/S earch match condition Finally, as seen in Figure 3 4, a P(BACK) spike was observed at Level 6 across all four conditions. This spike likely reflects careless decisions made by participants when they incorrectly navigated to Level 6 nodes and determined, by visual inspection of the forw ard preview nodes, that the target file was not present in that particular traversal path. In essence, instead of aiding performance, the presence of an additional decision making strategy at Level 6 (i.e., visual search) was associated with a degraded pe rformance across conditions. Moreover, the P(BACK) spike for t he memory/search no match condition was less prominent compared to all other conditions. This finding may again be attributed to an increased vigilance and carefulness among participants who w ere assigned to the no match condition. Re lating Speed and Accuracy of Node to Node Traversal Decisions Until this point, TON and P(BACK) have been used to represent the speed and accuracy of node based traversal decisions. However, as seen in Table 3 2, these two variables were not completely orthogonal to each other. Indeed, in line with speed accuracy tradeoffs observed in many cognitive studies, a 2 tailed Pearson r analysis showed that they shared a significant, positive correlation ( r = .13, p < .01 ). The statistical significance of this relationship was driven exclusively by the Memory/Search match condition ( r = .32, p < .01). These findings suggest that accuracy does not suffer as a result of more rapid decision making unless users are retrievin g files that they previously stored in a logical, hierarchical information structure. Perhaps knowing what one is searching for and having a relatively predictable environment in which to perform that search optimizes the extent to which speed and

PAGE 72

72 accurac y are log rolled. This, in turn, would give users a fairly binary decision in terms of how quickly or how accurately they make a decision. In the nonsensical tree (Memory condition), for example, the speed accuracy tradeoff was completely nonexistent. I n fact, the trend was actually in the opposite direction; a non significant negative correlation existed between TON and P(BACK) ( r = .06, ns ). This is not very surprising given the profuse degree of navigational uncertainties associated with that partic ular tree and condition. The fact that this speed accuracy correlation did not hold for the Memory/Search no match condition is somewhat bewildering. It was as if unfamiliar to be betwee In terms of tree level, the speed accuracy tradeoff was evident, but again only in a condition specific manner. When level and condition are considered, the relationship between TON and P(BACK) becomes quite intricate. In reviewing Figures 3 3 and 3 4, TONs decreased and P(BACK) executions increased (i.e., decisions were made faster but less accurately) as users d not occur within the same condition. TONs decreased with level depth in the Search condition, while P(BACK) increased with level depth in the Memory condition. Given that the Memory and Search conditions necessitated the use of two different cognitive strategies, one must be across levels. Task Difficulty Measures Table 3 3 presents descriptive statistics for variables related to task difficulty for each of the locations other than the ones pre established and anticipated by the experimenter.

PAGE 73

73 An independent samples t test revealed that Memory/Search match and Memory/Search t (22) = .90, ns ). Six between subjects, one way ANOVAs were conducted on each of the remaining dependent measures in Table 3 4. A main effect of Condition was found with re gard to failed to be retrieved files ( F (2,34) = 42.83 MSE = 137.32 2 = .75 p < .01), retrieval moves ( F (2,34) = 31.39 MSE = 83953.01 2 = .69 p < .01), and retrieval time ( F (2,34) = 24.46 MSE = 128.62 2 = .22 p < .01). Tukey HSD post hoc tests showed that users in the Memory condition committed significantly mor e retrieval failures and retrieval moves, and spent a greater amount of time retrieving files, than participants in both of the Memory/Search conditions ( p < .01). For retrieval time, Search was also outperformed by both of the Memory/Search conditions ( p < .05). In terms of storage time, a main effect of Condition was also observed ( F (2,34) = 4.49 MSE = 19.58 2 = .22 p < .05). However, post hoc tests revealed that the only statistically significant difference was the Memory/Search no match group requiring less times to store files than the Memory group. No significant differences existed among conditions wi th regard to storage failures, ( F (2,34) = 1.33 MSE = 3.00 2 = .08 ns ) and storage moves ( F (2,34) = .84 MSE = 2441.84 2 = .05 ns) Many of these performance measures were inter correlated. For example, the number of ely correlated with total number of retrieval moves in both t he Memory/S earch match condition ( r = .76, p < .01) and t he Memory/S earch no match condition ( r = .67, p be val number of storage moves, total storage time, or total retrieval time. This may be due to a lack of statistical power, given that the trends were in the expected di rections. Correlation matrices

PAGE 74

74 indicating relationships between all storage and retrieval metrics, both overall and separated by condition, can be found in Tables 3 4, 3 5, 3 6, 3 7, and 3 8. Item Analyses Task difficulty also appears to have depended on the specific file which was to be stored and/or retrieved, as well as the serial position in which it was retrieved relative to other files. Retrieval sequence, which was randomized for each participant, is addressed in the Correlational Analyses section of the Results. In the present section, we focus on the finding that certain files proved more difficult to store or retrieve than others. way ANOVAs revealed main effects of File fo r each of these dependent measures: mean number of storage moves for a particular file ( F (15, 559) = 2.41, MSE = 113.27, 2 = .06, p < .01), mean storage time for a particular file ( F (15, 559) = 3.99, MSE = 1257.74, 2 = .10, p < .01), mean number of retrieval moves for a particular file ( F (15, 735) = 2.58, MSE = 227.98, 2 = .05, p < .01), and mean retrieval time for a part icular file ( F (15, 735) = 3.28, MSE = 1615.38, 2 = .06, p < .01). Tukey HSD post hoc tests showed significant file to file contrasts. As measured by [a Mast erCard statement], Shakira Entries.txt [high school diary entries] were the least difficult files to store (mean: ~9 moves) ( p < .05). On the other hand, Rehearsal Schedule.doc [a dance club rehearsal schedule] was the most difficult file to store (mean: ~15 moves). the least difficult files to store (mean: ~23 seconds), whereas Rehearsal Schedule.doc,

PAGE 75

75 recommendation for graduate school] were the most difficult files to store (mean: ~38 sec onds) ( p < .05). Although the post hoc tests revealed no significant contrasts with regard to retrieval Freshman Year Entries.txt, and References.doc [sources for a term paper] to be the least difficult files to retrieve (mean: ~11 moves) while Software Costs.doc [money spent on course packet software], Man: The Social Animal [a course assigned article], Rehearsal Schedule.doc, and Confirmation #.html were the most difficult files to retrieve (mean: ~17 moves). As measured by retrieval time, the least difficult files to retrieve were Shakira Lie and References.doc (mean: ~22 seconds), whereas Confirmation #.html and Receptionist Intern Posting.html [a n advertised internship] were the most difficult files to retrieve (mean: ~38 seconds). All of these file specific performance metrics are summarized in Table 3 9. Overall, it may be concluded that well defined files, i.e., file names more likely to lead to navigation decisions along a single traversal path (e.g., bill statements and music files), were the least difficult files to store and retrieve. On the other hand, more ambiguous files, i.e., file names more likely to provoke traversal decisions along more than one path, were the most difficult to store and retrieve. For example, the rehearsal schedule document may have been associated with a dance club at school or one joined by the PC user outside of school. Therefore, the decision to navigate into ambiguous. Practice Effects The order in which files were retrieved was the same as the order in which they were stored. This sequence was randomly generated for each participant (Of course, storage sequence did not apply to t he S earch baseline condition in which there was no file storage.) For

PAGE 76

76 t he Memory/S earch no match condition it applied insofar as retrieved files and their analogous stored files were administered (and act ed upon) in the same order. Analyses were conducted to determine whether or not practice effects applied to the file retrieval task. That is, did users improve at file retrieval after they had worked with more and more files and became more and more fam iliar with the tree? Pearson correlations were calculated for file serial position and each of the following variables: number of storage moves for that file, storage time for that file, number of retrieval moves for that file, and retrieval time for th at file. File serial position was only negatively correlated with time measures of storage and retrieval performance. This means that users were able to store and retrieve files more and more quickly as they stored and retrieved more files. These correl ations are summarized in Table 3 10. I was curious to know whether or not task performance obeyed the power law of practice. Therefore, the two axes in the following graphs which plot performance, as measured by storage time, retrieval time, storage mov es, and retrieval moves, as a function of file position were log transformed. The resulting plots (see Figure 3 5) were well fit by a linear function, indicating that task performance, particularly as measured by storage and retrieval time, obeyed the pow er law of practice. That is, performance times, as a function of file serial position. As users stored and retrieved more files (i.e., as they drew closer to storing/retrieving their sixteenth and final file), they spent less time storing and retrieving e ach subsequent file, presumably because their familiarity with the tree grew progressively. The number of storage and retrieval moves was not significantly related to serial position, although statistical trends were in the expected directions. These res ults are not the first to suggest that learning in an electronic environment obeys the power law of practice. Johnson, Bellman, and Lohse (2003)

PAGE 77

77 found that website usage efficiency resulting from learning could also be modeled with the power law of practi ce. They showed that most websites could be characterized by linearly decreasing log transformed visit times as the frequency with which those sites were visited (also log transformed) increased. From an e commerce perspective, those sites with the fast est learning curves generally exhibited the highest rates of purchasing. Individual Differences in Working Memory Capacity and Storage Retrieval Time Interval Individual working memory capacity was assessed using the OSPAN. The mean OSPAN score across a ll participants was 14.04, although OSPAN performance was quite variable (SD = 8.16). For a given participant, storage retrieval lag time was calculated as the time difference between first file storage completion and first file retrieval commencement. I n other words, lag time referred to the amount of time that transpired between when the participant finished storing the first file and when the participant started retrieving the first file. Correlations between OSPAN and task performance and between st orage retrieval lag time and task performance are summarized, overall and by condition, in Table 3 11. It is important to note that these data treat each variable as an independent predictor of task performance. Inter correlations between OSPAN and lag t ime are discussed later in this section. As seen in Table 3 11 and Figure 3 6, a significant negative correlation existed between OSPAN performance and a number of performance measures, including retrieval failures, retrieval moves, and retrieval times. T hat is, participants with greater working memory capacity outperformed lower span participants on the file retrieval task. This finding is only partially consistent with original predictions given the results of the hierarchical regression presented analy ses later in this section. The overall working memory capacity retrieval time relationship was driven by significant correlations between OSPAN score and retrieval time in t he Memory/S earch match condition ( r = .60, p < .05). This correlation was not si gnificant for any

PAGE 78

78 of the other three conditions, indicating that increased working memory capacity could only serve as a facilitator when search is augmented by memory. In addition, OSPAN performance did not significantly correlate with storage failures, storage moves, and storage times. This suggests that file retrieval is more sensitive than file storage to working memory capacity. As mentioned, storage retrieval lag time was calculated as the time difference between first file storage completion and f irst file retrieval commencement (i.e., the amount of time that transpired between when the participant finished storing the first file and when the participant started retrieving the first file). Like working memory span, storage retrieval lag time appea red to have stronger relationships with file retrieval performance than with file storage performance. This effect is illustrated in Figure 3 7. The time lag was significantly correlated with retrieval failures ( r = .44, p < .01) but not with storage fai lures. The lag was significantly associated with retrieval moves ( r = .46, p < .01) but not with storage moves. However, the lag was significantly related to both retrieval time and storage time. These results supported the original hypothesis that grea ter lag times would be associated with degraded file retrieval performance. Admittedly, these two variables of interest (i.e., working memory capacity and storage retrieval retention interval) were not able to fully or independently predict task performa nce. Indeed, individual working memory capacity as assessed using the OSPAN was negatively correlated with storage retrieval time lag ( r = .37, p < .05). Higher working memory span participants demonstrated shorter storage retrieval lags than lower span participants, presumably because higher span participants completed the OSPAN more quickly. Therefore, a step wise, hierarchical regression analysis was conducted to partial out the effects of each of these two variables on file retrieval performance. G iven that the storage phase of each experimental session comprised part of the storage retrieval lag time, no analyses utilizing storage based

PAGE 79

79 dependent measures are reported textually or graphically. Nevertheless, correlations between lag time and storag e performance measures (as well as retrieval based measures) are reported in Table 3 7. With regard to retrieval moves, storage retrieval lag time was more strongly correlated with task performance than OSPAN score. Therefore, lag time was entered into the regression model first and accounted for a statistically significant amount of the variance (21%) observed in retrieval moves ( R 2 = .21 p < .01). Once that variance was accounted for, entering OSPAN score into the model increased R 2 to .27, thus acco unting for an additional 6% of the variance. However, this incremental predictability was not statistically significant. In terms of retrieval time, storage retrieval lag time was again more strongly correlated with task performance than OSPAN score. The refore, lag time was entered into the regression model first and accounted for a statistically significant amount of the variance (41%) observed in retrieval time ( R 2 = .41 p < .01). Once the variance due to lag time was partialed out, entering OSPAN sco re into the model increased R 2 to .48, thus accounting for an additional 7% of the variance. This delta R 2 represented a statistically significant increase in predictability of retrieval time ( p < .05). It can be concluded from these findings that partic ipants possessing greater working memory spans performed better on the file retrieval tasks, thus supporting original hypotheses. Higher span participants may have been more adept than lower span participants at remembering which paths had already been se arched and which paths had not yet been explored. A more precise memory record of which operators had been applied and where, along with the outcome (i.e., success or failure) of those operator applications, may have ultimately translated into faster and more accurate decisions.

PAGE 80

80 Nonetheless, the results suggested that this relationship between working memory capacity and retrieval performance was partly moderated by a third variable file storage retrieval retention time that, somewhat unexpectedly predicte d retrieval performance even better. The predictive power of this variable was particularly surprising given its limited variability (i.e., restriction of range). Every participant received the same intermediate task (i.e., the OSPAN). Therefore, users experienced fairly comparable storage Perhaps an even stronger relationship between storage retrieval time and performance would have been observed if the time lag data did not exhibit such a restriction of range. Finally, OSPAN score might not have predicted file retrieval performance as well as anticipated because it was measuring symbolic memory (i.e., remembering words) to a much greater extent than spatial memory. Had a more visual spatial measure of working m emory been used, we might have witnessed greater predictive capabilities. Subjective Reports of Strategy Participants were asked to provide an overall estimate (out of 100%) of the degree to which they used memory versus search as the overall cognitive st rategy during file retrieval. The mean percentages ascribed to memory and search were 51.39% (25.94) and 48.61% (25.94), respectively. There was no difference between males and females with regard to these overall reported strategies ( t (16) = .65, ns ). However, in focusing on subjective reports of memory strategy, a significant finding regarding participant sex emerged ( t (16) = 2.50, p < .05). As seen in Figure 3 8, men (M = 55.00%, SD = 27.89) reported using more of a spatial memory strategy than women (M = 30.42%, SD = 18.02). On the other hand, women (M = 69.58%, SD = 18.02) were far more likely to report having used a symbolic memory strategy than men (M = 45.00%, SD = 27.89).

PAGE 81

81 These findings are consistent with the literature reporting a preferentia l advantage for men in the processing and recall of visual spatial information and a preferential advantage for women in the processing and recall of symbolic, verbal information (e.g., Maccoby & Jacklin, 1974). The present study suggests that this dissoc iation persists beyond performance into the meta knowledge used when one assesses his or her own memory strategies. Given these subjective report data, the question that logically follows is: Was self reported strategy related to overall performance? The evidence overwhelmingly suggests that it is not. None of the performance measures (i.e., storage moves, storage time, retrieval moves, retrieval time) were significantly correlated with the overall self reported strategy (i.e., memory vs. search) or with the specific self reported memory strategies (i.e., spatial vs. symbolic). These analyses are depicted in Table 3 12 and Figures 3 9, 3 10, 3 11, and 3 12. Keep in mind that memory and search proportions added to 1, as did spatial memory and symbolic m emory proportions. That is, Percentage Search = 100% Percentage Memory and Percentage Symbolic Memory = 100% Percentage Spatial Memory. Therefore, only one member from each of these two pairs is graphed on horizontal axes in these figures (i.e.., Mem ory % for the self reported overall strategy in Figures 3 9 and 3 10; Spatial % for the self reported memory strategy in Figures 3 11 and 3 12. Using the other member on the horizontal axes (i.e., Search % for the self reported overall strategy and Symbol ic % for the self reported memory strategy) would have merely resulted in mirror image scatter plots with identical regression slopes (just a different sign). In short, self reported strategy neither facilitated nor impeded task performance. This is parti cularly surprising given the strength of the relationship between subjective strategy reports and other variables (i.e., sex of participant). Therefore, it can be concluded that self reports of

PAGE 82

82 file retrieval strategy merely reflect a personal preference or presumed approach and are not related to actual performance in either direction. This conclusion is consistent with the notion that retrospective, global evaluations are notoriously unreliable (e.g., Ericsson & Simon, 1993). Semantic Organization of Fo lder Names Experiment 1 generated two trees populated with folder names, relatively equated in terms of semantic relatedness and appropriately disparate with regard to super /subordinate structure. However, on a node by node basis, semantic relatedness an d super/sub relationships were not balanced within or between the two trees. A pair wise, correlational analysis was conducted to determine if these variables were related to file retrieval performance. The two mean ratings for each folder name pair in E xperiment 1 were aligned with the mean TON data for each of those pairs obtained in Experiment 2. Results demonstrated a significant, negative correlation between mean TON and mean relatedness score ( r = .18, p < .05). Follow up analyses confirmed that this relationship was significant for Tree A ( r = .23, p < .05) but not for Tree B ( r = .16, ns ). No significant correlations between TON and super/sub scores were obtained. In addition, no significant correlations were found between P(BACK) and either of the two score types for either tree. These correlations are summarized in Table 3 13 and illustrated graphically for the Log TON variable in Figure 3 13 and for the P(BACK) variable in Figure 3 14. Consistent with original hypotheses, these data sugge st that traversal decisions are facilitated at tree segments where semantic distances between nodes are minimized. However, this facilitation exhibits two types of specificities. First, it appears to enhance performance as measured by time but not by acc uracy. Second, it seems to be specific to an information environment in which users can use both reasoned, logical comparisons (search) and episodic traces (memory). Not surprisingly, closely related folder names seem to lose their ministering effect in nonsensical, illogical information structures (e.g., Tree B).

PAGE 83

83 Table 3 1. Mean Relatedness and Super/Sub scores and inclusion criteria for Tree A and Tree B folder names. Tree Mean Relatedness score Mean Super/Sub score Semantic Relatedness criteria Super /Sub criteria A 3.47 (.43) 3.52 (.47) 1.30 < Z SR < +1.30 Z SS > + .30 B 1.88 (.79) .90 (.73) 1.30 < Z SR < +1.30 Z SS < .35 Figure 3 1. Comparison of node based decision times by condition. Figure 3 2. Comparison of backup frequencies by condition

PAGE 84

84 Figure 3 3. Comparison of node based decision times by condition and tree level. Figure 3 4. Comparison of backup frequencies by condition and tree level.

PAGE 85

85 Table 3 2. Log TON P(BACK) correlations, overall and by condition. Condition Log TON P(BAC K) correlation Overall .13** Memory .06 Search .02 Memory/Search match .32** Memory/Search no match .01 *. Correlation is significant at the .05 alpha level (2 tailed). **. Correlation is significant at the .01 alpha level (2 tailed). Table 3 3. Performance metrics and values for OSPAN, storage, and retrieval by condition. Performance measure Mean (standard deviation) by condition Memory: N/A Search: N/A Memory/S earch match: 3.58 files (3.48) Memory/S earch no match: 2.50 files (2.28) Number files failed to be stored due to click/time restrictions Memory: 1.55 files (2.21) Search: N/A Memory/S earch match: .75 files (1.06) Memory/S earch no match: .58 files (1.00) Number files failed to be retrieved due to click/time restrictions Memory: 8.45 files (2.51) Search: 2.00 files (1.48) Memory/S earch match: 1.75 files (1.87) Memory/S earch no match: .83 files (1.03) Total number of storage moves (all 16 files) Memory: 195.45 moves (86.08) Search: N/A Memory/S earch match: 174.17 moves (33.02) Memory/S earch no match: 167.33 moves (25.20) Total number of retrieval moves (all 16 files) Memory: 347.64 moves (65.06) Search: 205.09 moves (52.49) Memory/S earch match: 176.83 moves (54.67) Memory/S earch no match: 156.83 moves (29.52) Total storage time (all 16 files) Memory: 9.42 minutes (2.43) Search: N/A Memory/S earch match: 7.53 minutes (2.29) Memory/S earch no match: 6.90 minutes (1.43) Total retrieval time (all 16 files) Memor y: 12.77 minutes (3.29) Search: 8.30 minutes (1.99) Memory/S earch match: 5.66 minutes (2.35) Memory/S earch no match: 5.60 minutes (1.05)

PAGE 86

86 Table 3 4. Correlation matrix showing storage and retrieval metrics across all conditions S torage failures Storage moves Storage time Retrieval failures Retrieval moves Retrieval time 1 .15 .14 .42** .87** .87** .83** Storage failures .15 1 .81** .49** .40* .45** .26 Storage moves .14 .81** 1 .67** .28 .35* .16 Storage time .4 2** .49** .67** 1 .42* .47** .55** Retrieval failures .87** .35* .28 .42* 1 .98** .83** Retrieval moves .87** .45** .35** .47** .98** 1 .85** Retrieval time .83** .26 .16 .55* .83* .85** 1 *. Correlation is significant at the .05 alpha level (2 tailed) **. Correlation is significant at the .01 alpha level (2 tailed). Table 3 5. Correlation matrix showing storage and retrieval metrics for Memory condition. Storage failures Storage moves Storage time Retrieval failures Retrieval moves Retrieval time 1 .92** .85** .32 .08 .20 .39 Storage failures .92** 1 .86** .31 .35 .48 .12 Storage moves .85** .86** 1 .67* .12 .26 .24 Storage time .32 .31 .67* 1 .20 .08 .08 Retrieval failures .08 .35 .12 .20 1 .96** 58 Retrieval moves .20 .48 .26 .08 .96** 1 .56 Retrieval time .39 .12 .24 .08 .58 .56 1 *. Correlation is significant at the .05 alpha level (2 tailed). **. Correlation is significant at the .01 alpha level (2 tailed).

PAGE 87

87 Table 3 6. Correlation matr ix showing storage and retrieval metrics for Search condition. Storage failures Storage moves Storage time Retrieval failures Retrieval moves Retrieval time N/A N/A N/A N/A N/A N/A N/A Storage failures N/A N/A N/A N/A N/A N /A N/A Storage moves N/A N/A N/A N/A N/A N/A N/A Storage time N/A N/A N/A N/A N/A N/A N/A Retrieval failures N/A N/A N/A N/A 1 .91** .25 Retrieval moves N/A N/A N/A N/A .91** 1 .29 Retrieval time N/A N/A N/A N/A .25 .29 1 *. Correlation is significan t at the .05 alpha level (2 tailed). **. Correlation is significant at the .01 alpha level (2 tailed). Table 3 7. Correlation matrix showing storage and retrieval metrics for Memory/Search match condition. Storage failures Storage moves Storage time Retrieval failures Retrieval moves Retrieval time 1 .14 .55 .26 .63* .76** .44 Storage failures .14 1 .80** .65* .24 .36 .34 Storage moves .55 .80** 1 .74** .46 .67* .54 Storage time .26 .65* .74** 1 .31 .44 .74** Retrieval failures .63* .24 .49 .31 1 .95** .72** Retrieval moves .76** .36 .67* .44 .95** 1 .79** Retrieval time .44 .34 .54 .74** .72** .79** 1 *. Correlation is significant at the .05 alpha level (2 tailed). **. Correlation is significant at the .01 alpha lev el (2 tailed).

PAGE 88

88 Table 3 8. Correlation matrix showing storage and retrieval metrics for Memory/Search no match condition. Storage failures Storage moves Storage time Retrieval failures Retrieval moves Retrieval time 1 .14 .01 .12 .81** .67* .55 Storage failures .14 1 .48 .56 .46 .45 .59* Storage moves .01 .48 1 .87** .03 .09 .28 Storage time .12 .56 .87** 1 .01 .07 .39 Retrieval failures .81** .46 .03 .01 1 .92** .78** Retrieval moves .67* .45 .09 .07 .92** 1 .8 8** Retrieval time .55 .59* .28 .39 .78** .88** 1 *. Correlation is significant at the .05 alpha level (2 tailed). **. Correlation is significant at the .01 alpha level (2 tailed).

PAGE 89

89 Table 3 9. Names, descriptions, locations and mean performan ces associated with specific files. File name File description Tree location Mean performance 1 January This file is a copy of your January 2007 cell phone bill. (7, 6) Storage moves: 9.06 (5.10) Storage time: 22.72 sec (14.62) Retrieval moves: 11.91 (9.00) Retrieval time: 23.98 sec (19.95) Statement.pdf This file is a copy of your December 2006 cell phone bill. (7, 5) 2 Statement.pdf This file is a copy of your MasterCard September 2006 statement. (7 12) Storage moves: 8.66 (4.89) Storage time: 22.97 sec (14.30) Retrieval moves: 10.61 (7.68) Retrieval time: 24.59 sec (21.92) Payment History.doc This file lists all of your payments of previous MasterCard balances. (7, 11 ) 3 Software Cost s.doc This file is a list of the money you spent on software that accompanies some course packets. (7, 24) Storage moves: 11.17 (6.94) Storage time: 28.69 sec (15.76) Retrieval moves: 16.59 (10.03) Retrieval time: 35.71 sec (22.74) Packet Costs.doc T his file is a list of the money you spent on course packets. (7, 23) 4 Payment Receipt.jpg This file is a copy of the payment receipt for your on campus parking pass. (7, 26) Storage moves: 12.49 (6.78) Storage time: 30.48 sec (15.69) Retrieval mo ves: 12.74 (8.80) Retrieval time: 26.79 sec (19.68) Citation Total.doc This file contains the amount of money that you owe for on campus parking citations. (7, 25) 5 Key West This file is a picture of you and your friends in Key West duri ng Spring Break 2005. (7, 39) Storage moves: 9.74 (6.64) Storage time: 21.65 sec (13.76) Retrieval moves: 13.22 (9.76) Retrieval time: 25.53 sec (19.79) Bahamas This file is a picture of you and your friends in the Bahamas during Spring Break 2 004. (7, 40)

PAGE 90

90 Table 3 9. Continued File name File description Tree location Mean performance 6 Shakira Hips This file is an MP3 audio file. (7, 41) Storage moves: 8.83 (5.37) Storage time: 20.98 sec (12.67) Retrieval moves: 11 .07 (8.93) Retrieval time: 22.21 sec (18.47) Britney Spears Everytime Lyrics.html This file is an MP3 audio file. (7, 42) 7 Freshman Year Entries.txt This file contains diary entries from your freshman year of high school. (7, 53) Storage moves: 9.69 (4.00) Storage time: 25.32 sec (14.16) Retrieval moves: 11.43 (8.30) Retrieval time: 25.06 sec (22.76) Sophomore Year Entries.text This file contains diary entries from your sophomore year of high school. (7, 54) 8 James Joyce Clay.pdf Thi s file is a short story for recreational reading. (7, 59) Storage moves: 11.51 (8.25) Storage time: 32.15 sec (20.53) Retrieval moves: 13.48 (9.80) Retrieval time: 26.82 sec (20.69) Nikolai Gogol The Cloak.pdf This file is a short story for recreationa l reading. (7, 60) 9 References. Doc This file contains the references for your Behavioral Analysis term paper. (7, 70) Storage moves: 10.71 (6.56) Storage time: 29.44 sec (17.79) Retrieval moves: 10.93 (7.40) Retrieval time: 23.80 sec (16.66) A bstract.doc This file contains the abstract (introduction) for your Behavioral Analysis term paper. (7, 69) 10 Man: The Social Animal.pdf This file is an article that was assigned in one of your classes. (7, 76) Storage moves: 11.00 (7.34) Storage t ime: 29.73 sec (14.13) Retrieval moves: 16.28 (10.92) Retrieval time: 36.12 sec (24.06) Skeletal Biological Distances.pdf This file is an article that was assigned in one of your classes. (7, 75)

PAGE 91

91 Table 3 9. Continued File name File description T ree location Mean performance 11 Psi Chi Requirements. html This file lists the requirements to become a member of this psychology honors society. (7, 88) Storage moves: 10.77 (6.61) Storage time: 28.35 sec (15.41) Retrieval moves: 12.74 (8.64) Ret rieval time: 28.71 sec (20.47) Pi Gammu Mu Regional Members.xls This file lists the members of this social science honors society that you are in. (7, 87) 12 Rehearsal Schedule.doc This file contains the rehearsal schedule for the Dance Club that you are in. (7, 90) Storage moves: 15.34 (8.37) Storage time: 39.38 sec (22.77) Retrieval moves: 16.87 (9.96) Retrieval time: 35.98 sec (25.68) Members List.doc This file lists the members of the Dance Club that you are in. (7, 89) 13 Confirma tion #.html This file contains the confirmation number provided to you when you registered for the GRE General. (7, 103) Storage moves: 12.54 (6.81) Storage time: 37.53 sec (23.51) Retrieval moves: 16.52 (10.66) Retrieval time: 38.22 sec (26.27) Test Location.html This file contains the test location provided to you when you registered for the GRE General. (7, 104) 14 Letter.doc This file is a letter written by Dr. White recommending you for graduate school. (7, 105) Storage moves : 11.06 (6.57) Storage time: 30.81 sec (19.44) Retrieval moves: 13.87 (8.96) Retrieval time: 30.28 sec (19.32) Letter This file is a letter written by Dr. Fischler recommending you for graduate school. (7, 106) 15 FL License Requi rements. html This file lists the state license requirements for becoming a real estate agent in Florida. (7, 117) Storage moves: 12.94 (8.85) Storage time: 35.21 sec (22.10) Retrieval moves: 15.04 (10.51) Retrieval time: 35.03 sec (25.43) Real Estate Courses List.html This file lists the course requirements for becoming a real estate agent in Florida. (7, 118)

PAGE 92

92 Table 3 9. Continued File name File description Tree location Mean performance 16 Receptionist Intern Posting.html This file i s a posting for a paid internship as a receptionist in a psychologist's private practice. (7, 123) Storage moves: 13.00 (8.54) Storage time: 39.13 sec (21.64) Retrieval moves: 15.91 (10.11) Retrieval time: 39.84 sec (27.96) Psycho metrician Intern Post ing.html This file is a posting for a paid internship as a test administrator in a psychologist's private practice. (7, 124) [Number]. File retrieved in all conditions; file stored in memory & M/S match conditions. *. File stored in M/S no match conditio n.

PAGE 93

93 Table 3 10. Correlation matrix showing serial position of file and performance measures. Performance measure Serial position Storage time .23** Retrieval time .18** Storage moves .07 Retrieval moves .07 *. Correlation is significant at the .05 alpha level (2 tailed). **. Correlation is significant at the .01 alpha level (2 tailed). Figure 3 5. Scatter plots and regression lines for log file serial position versus task performance measures. A) Vs log storage time. B) Vs log retrieval t ime. C) Vs log storage moves. D) Vs log retrieval moves.

PAGE 94

94 Table 3 11. Correlation matrix showing OSPAN score, storage retrieval lag time, and performance measures (overall and by condition). Performance measure Condition OSPAN score Storage retrieval la g time Storage failures Overall .16 .27 Memory .22 .92** Search N/A N/A M/S match .12 .14 M/S no match .03 .14 Retrieval failures Overall .33* .44** Memory .30 .08 Search .00 N/A M/S match .41 .63* M/S no match .24 .81** Storage moves Overall .09 .29 Memory .21 .85** Search N/A N/A M/S match .15 .55 M/S no match .35 .01 Storage time Overall .30 .76** Memory .24 .32 Search N/A N/A M/S match .48 .26 M/S no match .06 .12 Retrieval moves Overall .34* .46* Memory .27 .20 Search .01 N/A M/S match .35 .76** M/S no match .37 .67* Retrieval time Overall .45** .64** Memory .43 .39 Search .26 N/A M/S match .60* .44 M/S no match .32 .55 *. Correlation is significant at the .05 alpha l evel (2 tailed). **. Correlation is significant at the .01 alpha level (2 tailed).

PAGE 95

95 Figure 3 6. Scatter plots and regression lines for log OSPAN score versus task performance measures. A) Vs retrieval failures. B) Vs retrieval moves. C) Vs retrieval t ime (all conditions). D) Vs retrieval time (Memory/Search match condition). Figure 3 7. Scatter plots and regression lines for storage retrieval lag time and task performance measures. A) Vs storage failures. B) Vs retrieval failures. C) Vs storage moves. D) Vs retrieval moves.

PAGE 96

96 Figure 3 8. Double dissociation between self reported memory strategy and sex. Table 3 12. Correlation matrix showing self reported strategies and performance measures and participant sex. Self reported overall s trate gy Self reported memory s trategy Memory Search Spatial Symbolic Storage moves .09 .09 .14 .14 Storage time .03 .03 .08 .08 Retrieval moves .24 .24 .08 .08 Retrieval time .23 .23 .17 .17 Participant sex .16 .16 .49* .49* *. Correlation is significant at the .05 alpha level (2 tailed). **. Correlation is significant at the .01 alpha level (2 tailed).

PAGE 97

97 Figure 3 9. Scatter plots and regression lines for self reported overall strategy (Memory %) and file storage performance measures. A) Rep orted memory % vs storage moves. B) Reported memory % vs storage time. Figure 3 10. Scatter plots and regression lines for self reported overall strategy (Memory %) and file retrieval performance measures. A) Reported memory % vs retrieval moves. B) Re ported memory % vs retrieval time.

PAGE 98

98 Figure 3 11. Scatter plots and regression lines for self reported memory strategy (Spatial %) and file storage performance measures. A) Reported spatial % vs storage moves. B) Reported spatial % vs storage time. Figu re 3 12. Scatter plots and regression lines for self reported memory strategy (Spatial %) and file retrieval performance measures. A) Reported spatial % vs retrieval moves. B) Reported spatial % vs retrieval time.

PAGE 99

99 Table 3 13. Correlation matrix showing f older name ratings and performance measures by tree. Performance measure Tree Semantic score Super/Sub score Log TON A .23* .15 B .16 .20 P(BACK) A .01 .01 B .01 .01 *. Correlation is significant at the .05 alpha level (2 tailed). **. Correla tion is significant at the .01 alpha level (2 tailed). Figure 3 13. Scatter plots and regression lines for mean TON versus folder name pair score. A) Vs mean Relatedness score (Tree A). B) Vs mean Relatedness score (Tree B). C) Vs mean Super/Sub s core (Tree A). D) Vs mean Super/Sub score (Tree B).

PAGE 100

100 Figure 3 14. Scatter plots and regression lines for mean P(BACK) versus folder name pair score. A) Vs mean Relatedness score (Tree A). B) Vs mean Relatedness (Tree B). C) Vs mean Super/Sub (Tree A) D) Vs mean Super/Sub (Tree B).

PAGE 101

101 CHAPTER 4 GENERAL DISCUSSION versu et al comparisons all refer to dichotomies between similar, if not identical, psy chological constructs. I wanted to investigate how these cognitive strategies underlie folder path traversal decisions in a PC based file retrieval task. W ithout the use of a file exploration tool, file retrieval entails to varying degrees two navigati on strategies: memory and search. The memory strategy accesses an episodic memory trace (acquired during prior interaction with the directory tree structure), indicating the direction or path in which the user must traverse in order to move closer to the target location (i.e., file of interest) The search strategy refers to a presumably more effortful process involving reasoned comparisons based on semantic information (i.e., hierarchically organized folder names). In short, navigating toward a file em bedded within a user derived, hierarchically organized directory structure entails independent contributions of these two distinct problem solving strategies, one mediated by episodic memory, the other mediated by semantic memory. As Reder (1982) acknowle dged, both strategies consist of two major time consuming stages: 1) searching for needed information and 2) evaluating the adequacy of the information that was retrieved [from episodic or semantic memory]. The central goal of this research was to determi ne whether memory or search held a preferential advantage over the other strategy during file retrieval (baseline objective). An equally important objective was to determine the relative contributions of memory and search to individual navigation decision s inside the binary tree (experimental objective). Findings related

PAGE 102

102 to these primary goals, as well as to other, secondary objectives, are discussed in terms of original hypotheses (i.e., support or non support of initial predictions) and what was actuall y found, with explanations based on existing literature. The General Discussion is organized such that it first addresses each of the major findings that warrant discussion above and beyond that in the Results and Discussion. Next, implications of the re sults are addressed in the context of PC based file/folder exploration applications. Finally, future research directions are discussed. Summary of Findings Population of Binary Trees (Experiment 1) The purpose of this preliminary experiment was to populat e the nodes of two binary trees with folder and file names. Specifically, Tree A was to contain folder names that, pair wise, possessed relatively high super/sub scores, while Tree B was to include folder names that, pair wise, had relatively low super/su b scores. In this way, Tree A would make logical, intuitive, hierarchical sense, while Tree B would be illogical and nonsensical. While this objective was achieved, there were sizeable differences between the Relatedness of Tree A and the Relatedness of Tree B. The mean Relatedness scores for Tree A and Tree B were 3.47 ( SD = .43) and 1.88 ( SD = .79), respectively. Given the inherent correlation between Super/Sub characteristics and Semantic Relatedness, it was admittedly extremely difficult, if not im possible, to select folder names that were hierarchically related but that are no more semantically similar to each other than folder names that are not hierarchically related. Although the selected folder and file names met pre established inclusion crit eria, the Relatedness disparities between trees may have accounted for some performance differences. Indeed, as seen in Figure 3 13, Log TON was negatively correlated with the Relatedness of Tree A. That is, semantically similar folder names in Tree A we re associated with faster decisions at those nodes.

PAGE 103

103 Speed of Memory and Search Baseline Traversal Decisions (Experiment 2) Initially, a variety of hypotheses based on existing literature were made regarding memory memory is very good, for example at very short delays after acquisition. In my experiment, I predicted th at episodic memory traces would also be very strong for decisions at and near the root and leaf nodes, akin to a primacy and a recency effect. Specifically, it was hypothesized that search would yield faster decisions than memory at all nodes except those at and proximal to the root and leaf nodes, where episodic memory traces of paths traversed during storage were presumably very strong. This prediction received partial empirical support. Speed wise, Memory outperformed Search but only at shallow levels of the tree (i.e., root node, Level 1, and Level 2). However, no such Memory advantage was observed at deeper tree levels. Perhaps the strength of episodic memory traces was greater at deeper levels because they were not as frequently visited as more sh allow levels. An alternative explanation also exists. Figures 3 3 graphically illustrates the marked difference between Memory and Search TONs across tree levels. Excluding Levels 0 and 7, Search TONs decreased steadily into deeper levels, while Memory TONs remained relatively stable. A search specific funnel effect may exist, whereby the more specific folder names of deeper tree levels facilitated Search decisions (but not Memory decisions) relative to the more general folder names of shallow tree lev els. Perhaps, more specific folder names pruned semantic memory search space but not episodic memory search space, resulting in the morphological differences between the Search and Memory curves in Figure 3 3.

PAGE 104

104 Accuracy of Memory and Search Baseline Trave rsal Decisions (Experiment 2) As seen in Figure 3 4, the accuracy of memory was abysmal, particularly at deeper tree levels. In the Results and Discussion, a general overconfidence in memory was offered as one possible explanation. However, what gave par ticipants this unsubstantiated confidence in the path when in fact they did not know it? Several researchers (e.g., Schunn et al. 1997; Reder & Ritter, 1992) ha ve investigated these questions, using primarily arithmetic problems as their test stimuli. Arithmetic was chosen as the problem solving domain because it permitted rapid learning over multiple exposures and allowed the experimenters to independently vary the frequency of exposure to (and knowledge of) the problems and problem parts. The general paradigm was such that participants were trained on novel two digit x two digit math problems (e.g., 39 x 42). Before answering, participants were asked to make a quick feeling of knowing (FOK) judgment as to whether they could directly retrieve the answer from memory or had to calculate it. Somewhat surprisingly, FOK was better predicted by the frequency of presentation of the problem parts than by knowledge of answers to complete problems. This finding implies that knowing an answer represents a very different psychological construct than the feeling of knowing an answer and that these two constructs are likely mediated by different underlying cognitive process es. Choice models, like the one proposed by Reder and Ritter (1992), posit that strategy choice reflects level of FOK. In the context of the present experiments, participants may have actively and voluntarily selected a strategy based on their degre e of FOK. According this model, a high FOK would indicate that a particular trial would be best completed by relying on memory. On the other hand, a low FOK would suggest that the participant depend on search. While interacting with the tree, participan ts may have developed high FOKs with regard to the spatial

PAGE 105

105 and symbolic relationships of folder names when in fact they had no such knowledge. Alternatively, participants may have incorrectly recognized a folder name as familiar merely because it was spat ially near or symbolically (semantically) similar to another folder (name) that they had previously visited. These unfounded FOK judgments during file retrieval may account for the disproportionately high number of false positive observed in memory based traversal decisions. Memory/Search Hybrid Performance (Experiment 2) It was initially hypothesized that, even after tree level was accounted for, the speed and accuracy of the experimental groups would fall somewhere between that of the Memory and Search c onditions, reflecting a mixture of the two strategies. As is often the case with exploratory research, the original predictions were not supported Instead, Memory/Search hybrid performance was both faster and more accurate than either strategy alone. T herefore, it was impossible to conduct any analysis or to test any original hypothesis that assumed that memory/search values to each node/level within the tree). Table 4 1 summarizes several hypothetical findings of Experiment 2 with regard to the speed of node level traversal decisions. Theoretical interpretations of those findings are provided. In short, while potential finding #3 was what was hypothesized, potential fin ding #4 was what was actually observed. Table 4 2 summarizes several hypothetical findings of Experiment 2 with regard to the accuracy of node level traversal decisions. Theoretical interpretations of those findings are again supplied. Once more, while potential finding #3 was what was hypothesized, potential finding #4 was what was actually observed. The better than baseline performance of Memory/Search could possibly be explained in terms of parallel horse race models. Horse race models have tradition ally been used to describe

PAGE 106

106 the process by which category exemplars are retrieved from memory. However, such models may apply to the present research as well. Nosofsky and Palmeri (1997) used a horse race model to describe how participants performed Garne illustrated schematically in Figure 4 1. In this task, there are four stimuli varying along two dimensions, with two values per dimension. These dimension values are approximately equally discriminable. T he objective is to classify each stimulus into its assigned category as quickly and accurately as possible. A fundamental prediction of parallel horse race models is that the greater the number of tokens, or instances, of an exemplar that race to be retr time tends to be (e.g., Logan, 1988; Raab, 1962; Townsend & Ashby, 1962). Analogously, the more horses that participate in a race, the greater the probability is that at least one of the finish times will be parti of faster individual processes. A second prediction of horse race models is that experience with a particular exemplar increases the number of tokens of that exemplar that are stored in memory, thus facilitating performance on those particular items. As a result, when an item is presented at test, a greater number of tokens of the exemplar race to be retrieved. All activated exemplars race stochastically (yet, exponentially) to be retrieved from memory at a rate that can be mathematically determined by their activation values. If an analogy is drawn between exemplar tokens in the horse race model and navigational cues in the current study, then it is conceivable that the grea ter number of traversal cues in the Memory/Search conditions would yield a super additive performance (that is, the traversal decisions that are both faster and more accurate than those predicted by either baseline condition alone). In the Memory conditio

PAGE 107

107 of the current location folder name presented to the participant after a previous traversal mo ve was completed. In the Search condition, instances of category exemplars (related to folder names) may compete to be retrieved from semantic memory. In the Memory/Search condition, it were racing to be retrieved from the two memory systems simultaneously. This would explain the super additive performance of the hybrid conditions, in which two cognitive strategies based on two separable memory systems were available. Speed Accuracy Relationships in Traversal Decisions As mentioned in the Results, accuracy does not suffer as a result of more rapid decision making unless users are retrieving files that they previously stored in a logical, hierarchical inf ormation structure (i.e., Tree A of the Memory/Search match condition). Perhaps knowing what one is searching for and having a relatively predictable environment in which to perform that search enforces the inverse relationship between speed and accuracy. In other words, participants must choose between a quick decision and an accurate decision in this type of environment. In the nonsensical tree (Memory condition), on the other hand, the speed accuracy tradeoff was completely nonexistent. In fact, the trend was actually in the opposite direction; a non significant negative correlation existed between TON and P(BACK) ( r = .06, ns ). This is not very surprising given the profuse degree of navigational uncertainties associated with that particular tree an d condition. The fact that this speed accuracy correlation did not hold for the Memory/Search no match condition is somewhat bewildering. It was as if unfamiliar to be ecision and

PAGE 108

108 A full speed accuracy tradeoff was not observed in the present study when tree level was accounted for. Such a tradeoff was evident, but only in a condition specific manner. TONs decreased and P(BACK) executions increa sed (i.e., decisions were made faster but less the Search condi tion, while P(BACK) increased with level depth in the Memory condition. Given that the Memory and Search conditions necessitated the use of two different cognitive etween speed and accuracy across levels. up their decisions with no loss in accuracy. This result is perfectly analogous with the present h based decisions became faster in deeper tree levels, while accuracy of those decisions were unaffected. Therefore, in both studies, it can be concluded that Search participants were performing sub optimally. This conclusion is further supported by the f act that participants in the Memory/Search hybrid conditions made much quicker traversal decisions than Search alone participants. Reder (1982) found the same speed accuracy pattern for recognition were flat across tree levels in the current experiment, a true comparison between the two studies was not possible in this regard. Memory accuracies did decrease in deeper tree levels, although that was probably because were turned back upon realizing that the target file was not found on that traversal path. Spatial, Symbolic, and Working Memory Anderson, 1996) pitting memory agai nst search has been the primary focus of the current

PAGE 109

109 research, another theoretical question involving memory alone has received secondary emphasis. In situations where participants relied more heavily on memory than search, exactly what type of memory wer e they accessing spatial memory or symbolic memory? Spatial memory refers to increased. Symbolic memory, on the other hand, refers to remembering that o ne folder name was related hierarchically to another folder name, and so forth. As reported in the Results and Discussion, retrospective subjective reports of memory strategy were largely dependent on the sex of the participant but were not predictive of actual file storage/retrieval performance. However, if those reported strategies did affect performance, would spatial memory or symbolic memory be the more efficient retrieval strategy? Dumais and Jones (1985) conducted a personal information managemen t (PIM) study that addressed this very question. They designed a paper based experiment to test the assumption that spatial memory is more effective than symbolic memory. Participants were asked to organize a series of items, using a combination of locat ion based (spatial) and name based (symbolic) filing systems. Over a series of retrieval tasks, spatial filing was seen to offer no benefit over symbolic filing. Furthermore, retrieval speeds for the location only strategy deteriorated significantly as m ore items were added. Based on their findings, Dumais and Jones suggested that spatial management should not replace symbolic techniques, but instead act as an adjunct, much as it does in modern desktop computers. Although the spatial symbolic winner rema ins unknown, the role of working memory in file retrieval is undeniable. Participants must keep a running record of where they have traversed and the outcome (success or failure) of their navigation moves. In many ways, the task is similar to Milton Brad

PAGE 110

110 turn over two cards, revealing pictures on the other side. If the pictures match, the player removes those cards from the grid and gets points. If the pictures do not match, th e player must flip back over both cards and attempt to remember the pictures associated with each of them for future flips. Similarity judgments and analogical mapping also implicate the role of working memory in the present research. These domains are also heavily reliant on working memory and may be requires the formation of either a symbolic or spatial representations (Medin, Goldstone, & Genter, 1 993). Similarly, analogical mapping requires participants to retain and integrate images in working memory in order to identify correspondences between visual scenes (Waltz, Lau, Grewal, & Holyoak, 2000). Imagine the following two scenes: visual scene # 1 a woman is seen receiving food from the Community Food Bank; visual scene 2 another woman is seen feeding a squirrel in her backyard. On the basis of attribute mapping, the woman in the first picture could be mapped to the woman in the second. However, on the basis of relational (analogical) mapping, the woman in the first picture could be mapped to the squirrel in the second picture, as each is a recipient of food. Although the tasks necessitate different processes, analogical mapping and file retriev al both involve executive working memory. Identifying correspondences requires spatial and/or symbolic analogical maps of visual scenes to be held and integrated in working memory. Similarly, symbolic and/or spatial maps were needed to keep track of prev iously traversed path segments in the present study. Implications and Applications Ecological Validity The ecological validity of the present research is a source of great pride. The experiments have employed real world folder and file names, allowed par ticipants to store their own files in

PAGE 111

111 lieu of doing the storage for them, and allowed users to freely traverse hierarchical trees as they sought to retrieve those files. However, this research is not without generalizability concerns and threats to exter nal validity. The interface was different than the Microsoft Windows environment to which most users are accustomed. Double clicking on a folder did not drill one level deeper into the was provided. In addition, PC users are not accustomed to time and mouse click limits and rarely interact with a perfectly balanced seven level binary tree (eight levels, if you count the root node) that requires them to choos e from exactly two folders for each traversal move. With the increasing number of search tools available to consumers, there is a growing trend for PC users to flatten their file/folder hierarchies (Boardman & Sasse, 2004). Therefore, the seven level bin ary tree may have lacked external validity in this regard as well. The present research also failed to consider that some PC users do not like storing documents that are currently or were recently in use. Perhaps, to maximize retrievability, users would have preferred to leave such files on their PC desktops until they were finished working with them before deciding to embed them somewhere in the hierarchical tree. Indeed, in a paper s of documents on information readily available. These piles were often organized according to contextual factors circumstances (e.g., an ongoing project), and time (e.g., frequency of use or age of document). Lastly, it is highly unlikely that users would store all of their files before needing to retrieve at least one of them. A likely real world scenario would show a more staggered,

PAGE 112

112 interleaved file storage and retrieval. Dix and Marshall (2003) investigated a more complex relationship between storage and retrieval. Specifically, did retrieval depend on the time at which website bookmarks were organized? Two significantly faster recall. Howe ver, the difference was not evident when users were tested a week later. Dix and Marshall suggested that the initial result may be due to the close proximity of the sorting and post test tasks. These findings should be kept in mind given the methods of t he present study. For example, the following two procedures in the current experiments should be considered: (1) the time delay between storage and retrieval was on the order of minutes (not days) and (2) browsing (searching) was always done after sortin g (storing) has been completed. A key criticism that can be leveled at such controlled studies is that they lack ecological validity. Although a number of interesting results are presented, the relevance of the findings to real world contexts can be que stioned. Dix and Marshall highlighted two limitations regarding the external validity of their own experiments: (1) the bookmarks were pre selected and (2) real life users may use a combination of during and after browsing sorting. Whittaker et al (200 0) has called for greater integration of field and controlled studies in PIM research in an attempt to balance these types of internal and external validity tradeoffs. PC Based File/Folder Exploration Tools Although it was not an objective of the present r esearch, conclusions drawn from these studies may have implications for the front end design or re design of PC file/folder navigation applications. That is, changing the user interfaces of such tools may yield a more efficient and enjoyable human compute r interaction.

PAGE 113

113 In the real world, speed would likely be considered more important than accuracy. There is really no cost associated with navigating backward, other than perhaps a slight decline in user happiness or satisfaction. Therefore, time require d to retrieve the file of interest (i.e., speed) is taken as the most important metric. It can be inferred that maximization of decision speed (or minimization of TON) depends on two factors: 1) the degree to which the user is familiar with the tree and 2) the level of depth within the tree where the user currently resides. In the context of the current experiment, these contingencies are encapsulated in the TON based Condition x Level interaction. Memory participants were essentially nave users intera cting with an unfamiliar tree, whereas Search participants were effectively experienced users interacting with a familiar tree. Given that Memory performance did not change across tree levels, navigation tools may best serve users in their current static forms (e.g., Microsoft File Explorer) but only if the contents of the tree is familiar to the user. However, if the tree is unfamiliar, it may be useful for based de cision speeds increase as a function of tree depth. For example, when the user is residing at deeper tree levels (since that is where traversal speeds were best served by the Search strategy), the folder with the name that is most related to the one hierarchically above it is larger than the rest. Folder names could be assigned and ranked according to super/sub scores on the back end of the tool using latent semantic analysis (LSA), for instan ce (see Landeaur, 1998 for an introduction to LSA). This morphed folder would not necessarily be the next folder along the correct traversal path, but it would work in favor of the human factors at play if it happened to be the next folder along that path

PAGE 114

114 once the tree had passed some pre set familiarity threshold, given that familiar information structures would not seem to benefit from it at any tree level (based on th e Memory TON data). The notion of a dynamically changing user interface is just one possibility of how PC based file/folder exploration tools could be re designed to exploit our knowledge of the human factors associated with file retrieval. A great de al of additional research must be conducted before the industry commits to such design modifications and claims the changes to be empirically based and data driven. Future Work Future studies may wish to determine the extent to which findings from the pres ent experiments retain their integrity. For example, the tree size and structure in the current experiments were held constant. It would be interesting to increase (and decrease) the number of ity in terms of its branching so that traversal decisions were not merely binary. The imposition of various costs might be another implementation costs of pr oblem solving operators. Imposing costs such as lockout times following errors and manipulating mouse click limitations, as well as the lag time between file storage and retrieval, all may have had profound effects on the results of the present study. F inally, subsequent experiments that stagger the storage and retrieval phases may be helpful in achieving greater ecological validity. In future studies, it would be interesting to determine whether a horse race model or a choice model better fits the data. based on FOK, or are they a result of navigational cues competing to be retrieved? Analyzing talk aloud verbal protocols might be one method to experimentally discriminate between the two mo dels. In this way, I could determine whether participants possessed a conscious awareness of

PAGE 115

115 the strategies used during the traversal decisions themselves. Another technique may be to employ a procedural paradigm similar to that of Reder and Ritter (1992 ), whereby I could ask participants their initial FOK associated with a particular problem state. Specifically, I could require a judgment as to whether their next traversal move could be directly retrieved from based logic and heuristics based on folder names. From a user interface perspective, the next logical step would be to actually design a functional file exploration tool that exploits the human factors findings obtained in the current research, as di scussed earlier. It would then be useful, both practically and theoretically, to determine whether a dynamically adaptive tool that accounts for these human factors would exhibit time saving benefits above and beyond those reaped when an existing file/fol der application is the ultimate goal, much more research is required to flesh out the details regarding

PAGE 116

116 Table 4 1. Potential traversal speed findings and their theoretical interpretations (Exp. 2). Potential finding Theoretical interpretation 1. M > ( t = S) Individual relied exclusively on search. ( M & S are additive) 2. ( t = M) > S Individual relied exclusively on memory. ( M & S are additive) 3. M > t > S Individual relied partially on search and partially on memory; the presence of both retrieval strategies degraded speed performance relative to search alone. ( M & S are additive) 4. M = S > t Individual re lied partially on search and partially on memory; the availability of both retrieval strategies enhanced speed performance relative to either baseline strategy alone. ( M & S are non additive) 5. t > M > S Individual relied partially on search and partial ly on memory; the presence of both retrieval strategies degraded speed performance relative to either baseline strategy alone. ( M & S are non additive) M: Memory baseline TONs; S: Search baseline TONs; t : observed hybrid TONs Table 4 2. Potential tra versal accuracy findings and their theoretical interpretations (Exp. 2). Potential finding Theoretical interpretation 1. M > ( e = S) Individual relied exclusively on search. ( M & S are additive) 2. ( e = M) > S Individual relied exclusively on memory. ( M & S are additive) 3. M > e > S Individual relied partially on search and partially on memory; the presence of both retrieval strategies degraded accuracy performance relative to search alone. ( M & S are additive) 4. M > S > e Individual relied part ially on search and partially on memory; the availability of both retrieval strategies enhanced accuracy performance relative to either baseline strategy alone. ( M & S are non additive) 5. e > M > S Individual relied partially on search and partially on memory; the presence of both retrieval strategies degraded accuracy performance relative to either baseline strategy alone. ( M & S are non additive) M: Memory baseline error rates; S: Search baseline error rates; e : observed hybrid error rates

PAGE 117

117 Figure 4 from Nosofsky, R.M. & Palmeri, T.J. 1997. (p. 1030).]

PAGE 118

118 APPENDIX A EXPERIMENT 2: TREE A FOLDER/FILE NAMES Stimuli(0, 1) = "All My Stuff" Stimuli(1, 1) = "Personal" Stimuli(1, 2) = "School and Professional" Stimuli(2, 1) = "Financial Stuff" Stimuli(2, 2) = "Social Stuff" Stimuli(2, 3) = "Pre Graduation Stuff" Stimuli(2, 4) = "Post Graduation Stuff" Stimuli(3, 1) = "Bills" Stimuli(3, 2) = "School Expenses" Stimuli(3, 3) = "Multimedia" Stimuli(3, 4) = "Recreational Documents (Text)" Stimuli(3, 5) = "Coursework" Stimuli(3, 6) = "Pre Graduation Extracurricular Activities" Stimuli(3, 7) = "Grad School Preparation" Stimuli(3, 8) = "Professional Preparation" Stimuli(4, 1) = "Utilities and Communication" Stimuli(4, 2) = "Credit Cards" Stimuli(4, 3) = "Course Related Costs" Stimuli(4, 4) = "Non Course Costs" Stimuli(4, 5) = "Pictures" Stimuli(4, 6) = "Music" Stimuli(4, 7) = "Private Stuff" Stimuli(4, 8) = "Recreational Documents for Reading" Stimuli(4, 9) = "Psychology Major Courses" Stim uli(4, 10) = "General Education Courses" Stimuli(4, 11) = "Academic" Stimuli(4, 12) = "Recreational" Stimuli(4, 13) = "GRE Stuff" Stimuli(4, 14) = "Grad School Applications" Stimuli(4, 15) = "Prospective Jobs" Stimuli(4, 16) = "Prospective Internships" Stimuli(5, 1) = "Electric (GRU) Bills" Stimuli(5, 2) = "Phone Bills" Stimuli(5, 3) = "MasterCard" Stimuli(5, 4) = "Discover" Stimuli(5, 5) = "Tuition and Fees" Sti muli(5, 6) = "Books and Supplies" Stimuli(5, 7) = "Transportation Expenses" Stimuli(5, 8) = "Food and Entertainment" Stimuli(5, 9) = "Family Pics" Stimuli(5, 10) = "Friends Pics" Stimuli(5, 11) = "Songs" Stim uli(5, 12) = "Music Videos" Stimuli(5, 13) = "Letters"

PAGE 119

119 Stimuli(5, 14) = "Journal Entries" Stimuli(5, 15) = "Literature (Poems and Short Stories)" Stimuli(5, 16) = "Non Literature" Stimuli(5, 17) = "Cognitive Psycholo gy" Stimuli(5, 18) = "Behavioral Analysis" Stimuli(5, 19) = "General Anthropology" Stimuli(5, 20) = "Statistics I" Stimuli(5, 21) = "Club Memberships" Stimuli(5, 22) = "Honor Societies" Stimuli(5, 23) = "Soci al Clubs" Stimuli(5, 24) = "Intramural Sports" Stimuli(5, 25) = "GRE Subject (Psychology)" Stimuli(5, 26) = "GRE General" Stimuli(5, 27) = "Supplementary Materials" Stimuli(5, 28) = "Application Forms" St imuli(5, 29) = "Psychology Related Jobs" Stimuli(5, 30) = "Non Psychology Jobs" Stimuli(5, 31) = "Paid Internships" Stimuli(5, 32) = "Unpaid Internships" Stimuli(6, 1) = "Electronic Funds Transfers (EFT)" Stimuli(6, 2) = "Energy Consumption Summaries" Stimuli(6, 3) = "Cell Phone (Verizon Wireless)" Stimuli(6, 4) = "Landline Phone (Verizon)" Stimuli(6, 5) = "Expected Credits" Stimuli(6, 6) = "Previous Balances" Stimuli(6, 7) = "C ashBack Bonus Redemptions" Stimuli(6, 8) = "Credit Line Increase" Stimuli(6, 9) = "Junior Year" Stimuli(6, 10) = "Senior Year" Stimuli(6, 11) = "Textbooks" Stimuli(6, 12) = "Course Packets and Software" Stimu li(6, 13) = "On Campus Parking" Stimuli(6, 14) = "Gas Expenses (By Semester)" Stimuli(6, 15) = "Leisure Activities" Stimuli(6, 16) = "On Campus Meals and Snacks" Stimuli(6, 17) = "Holiday Pics" Stimuli(6, 18) = "Vaca tion Pics" Stimuli(6, 19) = "Gator Games Pics" Stimuli(6, 20) = "Spring Break Pics" Stimuli(6, 21) = "Pop" Stimuli(6, 22) = "Rap" Stimuli(6, 23) = "Country" Stimuli(6, 24) = "Rock" Stimuli(6, 25) = "O ld Love Letters" Stimuli(6, 26) = "Family Letters" Stimuli(6, 27) = "High School Diaries"

PAGE 120

120 Stimuli(6, 28) = "Current Online Journal" Stimuli(6, 29) = "Poems" Stimuli(6, 30) = "Short Stories" Stimuli(6, 31) = Articles" Stimuli(6, 32) = "Recipes" Stimuli(6, 33) = "Lecture Notes" Stimuli(6, 34) = "Exam Reviews" Stimuli(6, 35) = "Term Paper" Stimuli(6, 36) = "Course Info" Stimuli(6, 37) = "Old Exams" Stimuli( 6, 38) = "Assigned Articles" Stimuli(6, 39) = "Problem Sets" Stimuli(6, 40) = "Final Project" Stimuli(6, 41) = "Psychology" Stimuli(6, 42) = "Non psychology" Stimuli(6, 43) = "General" Stimuli(6, 44) = "Socia l Science" Stimuli(6, 45) = "Dance Club" Stimuli(6, 46) = "Marching Band" Stimuli(6, 47) = "Basketball" Stimuli(6, 48) = "Flag Football" Stimuli(6, 49) = "Sign up Directions" Stimuli(6, 50) = "Study Materials Stimuli(6, 51) = "Prep Course" Stimuli(6, 52) = "Registration Materials" Stimuli(6, 53) = "Letters of Recommendation" Stimuli(6, 54) = "Transcripts" Stimuli(6, 55) = "Harvard University" Stimuli(6, 56) = "U niversity of South Florida" Stimuli(6, 57) = "Research Assistant" Stimuli(6, 58) = "School Psychology Assistant" Stimuli(6, 59) = "Real Estate Agent" Stimuli(6, 60) = "Animal Breeder" Stimuli(6, 61) = "Shands Childre n's Hospital" Stimuli(6, 62) = "Private Practice Internships" Stimuli(6, 63) = "V.A. Hospital" Stimuli(6, 64) = "Alachua County Crisis Center" Stimuli(7, 1) = "Signup.html" Stimuli(7, 2) = "Cancellation.html" Stimuli(7, 3) = "July '06 Consumption.doc" Stimuli(7, 4) = "August '06 Consumption. doc" Stimuli(7, 5) = "December '06 Statement.pdf" Stimuli(7, 6) = "January '07 Statement.pdf" Stimuli(7, 7) = "October '06 Statement.pdf" Stimuli(7, 8) = "November '06 Statement.pdf" Stimuli(7, 9) = "Laptop Return Receipt.jpg"

PAGE 121

121 Stimuli(7, 10) = "Watch Return Receipt.jpg" Stimuli(7, 11) = "Payment History.doc" Stimuli(7, 12) = "September '06 Statement.pdf Stimuli(7, 13) = "Olive Garden Redemption.html" Stimuli(7, 14) = "Foot Locker Redemption.html" Stimuli(7, 15) = "General Info.doc" Stimuli(7, 16) = "Petition.doc" Stimuli(7, 17) = "3rd Year Tuition.doc" Sti muli(7, 18) = "3rd Year Fees.doc" Stimuli(7, 19) = "4th Year Tuition.doc" Stimuli(7, 20) = "4th Year Fees.doc" Stimuli(7, 21) = "On Campus Bookstore Purchase.bmp" Stimuli(7, 22) = "Off Campus Bookstore Purchase.bmp" Stimuli(7, 23) = "Packet Costs.doc" Stimuli(7, 24) = "Software Costs.doc" Stimuli(7, 25) = "Citation Total.doc" Stimuli(7, 26) = "Payment Receipt.jpg" Stimuli(7, 27) = "Fall '06.txt" Stimuli(7, 28) = "Spring '07.txt" Stimuli(7, 29) = "Football Season Ticket Receipt.jpg" Stimuli(7, 30) = "Performance Show Ticket Prices.doc" Stimuli(7, 31) = "Declining Balance Account Summary.html" Stimuli(7, 32) = "Vending Account Balance.html" S timuli(7, 33) = "Christmas '06.jpg" Stimuli(7, 34) = "Thanksgiving '06.jpg" Stimuli(7, 35) = "Europe '05.bmp" Stimuli(7, 36) = "California '04.bmp" Stimuli(7, 37) = "Football '06.jpg" Stimuli(7, 38) = "Basketball '06 .jpg" Stimuli(7, 39) = "Key West '05.jpg" Stimuli(7, 40) = "Bahamas '04.jpg" Stimuli(7, 41) = "Shakira Hips Don't Lie.mp3" Stimuli(7, 42) = "Britney Spears Everytime Lyrics.html" Stimuli(7, 43) = "50 Cent Candy Shop.mp3" Stimuli(7, 44) = "Snoop Dog Drop It Like It's Hot.mp3" Stimuli(7, 45) = "LeAnn Rimes Blue.mp3" Stimuli(7, 46) = "Keith Urban You'll Think Of Me.mp3" Stimuli(7, 47) = "Green Day Time Of Your Life.mp3" Stimuli(7, 48) = "Beatles I Want To Hold Your Hand.mp3" Stimuli(7, 49) = "To Ex.doc" Stimuli(7, 50) = "From Ex.doc" Stimuli(7, 51) = "From Mom.doc" Stimuli(7, 52) = "To Dad.doc" Stimuli(7, 53) = "Freshman Year E ntries.txt" Stimuli(7, 54) = "Sophomore Year Entries.txt" Stimuli(7, 55) = "Entries Only Friends Can See.html"

PAGE 122

122 Stimuli(7, 56) = "Entries Only I Can See.html" Stimuli(7, 57) = "Robert Frost Stars.pdf" Stimuli(7, 58) = "Maya Angelou Still I Rise.pdf" Stimuli(7, 59) = "James Joyce Clay.pdf" Stimuli(7, 60) = "Nikolai Gogol The Cloak.pdf" Stimuli(7, 61) = "How To Be A Leader.html" Stimuli(7, 62) = "Eating Healthy.doc" Sti muli(7, 63) = "Blueberry Crepes.html" Stimuli(7, 64) = "Pumpkin Pie.html" Stimuli(7, 65) = "Working Memory.ppt" Stimuli(7, 66) = "Problem Solving.ppt" Stimuli(7, 67) = "Exam 2 Review.doc" Stimuli(7, 68) = "Final Exam Review.doc" Stimuli(7, 69) = "Abstract.doc" Stimuli(7, 70) = "References.doc" Stimuli(7, 71) = "Syllabus.doc" Stimuli(7, 72) = "Blog Comments.html" Stimuli(7, 73) = "Fall 2006 Exam 1.doc" Stimuli(7, 74) = "S pring 2005 Exam 2.doc" Stimuli(7, 75) = "Skeletal Biological Distances.pdf" Stimuli(7, 76) = "Man: The Social Animal.pdf" Stimuli(7, 77) = "Normal Distribution.doc" Stimuli(7, 78) = "Probability.doc" Stimuli(7, 79) = "Data Collection.xls" Stimuli(7, 80) = "Data Analysis.xls" Stimuli(7, 81) = "Psychology Club Members.txt" Stimuli(7, 82) = "Future Therapists Members.txt" Stimuli(7, 83) = "Civil Engineers Society Info.txt" Stimuli( 7, 84) = "Architecture Club Info.txt" Stimuli(7, 85) = "Golden Key Requirements.html" Stimuli(7, 86) = "Phi Beta Kappa Requirements.html" Stimuli(7, 87) = "Pi Gamma Mu Regional Members.xls" Stimuli(7, 88) = "Psi Chi Requirem ents.html" Stimuli(7, 89) = "Members List.doc" Stimuli(7, 90) = "Rehearsal Schedule.doc" Stimuli(7, 91) = "Practice Schedule.pdf" Stimuli(7, 92) = "Contacts.txt" Stimuli(7, 93) = "Roster.doc" Stimuli(7, 94) = "Game Schedule.pdf" Stimuli(7, 95) = "Rules and Regulations.html" Stimuli(7, 96) = "Playbook.ppt" Stimuli(7, 97) = "Step By Step Instructions.pdf" Stimuli(7, 98) = "Interactive Instructions.html" Stimuli(7, 99) = "H istory Of Psychology Outline.doc" Stimuli(7, 100) = "Famous Psychologists Outline.doc" Stimuli(7, 101) = "Quantitative Materials.ppt"

PAGE 123

123 Stimuli(7, 102) = "Verbal Materials.ppt" Stimuli(7, 103) = "Confirmation #.html" S timuli(7, 104) = "Test Location.html" Stimuli(7, 105) = "Dr. White's Letter.doc" Stimuli(7, 106) = "Dr. Fischler's Letter.doc" Stimuli(7, 107) = "Community College Transcript.pdf" Stimuli(7, 108) = "University of Florida Tra nscript.pdf" Stimuli(7, 109) = "Clinical Psych Program App.pdf" Stimuli(7, 110) = "Industrial Org Psych Program App.pdf" Stimuli(7, 111) = "Cognitive Psych Program App.pdf" Stimuli(7, 112) = "Behavioral Analysis Program App. pdf" Stimuli(7, 113) = "Participant Schedule.html" Stimuli(7, 114) = "Lab Contact Info.txt" Stimuli(7, 115) = "Middle School Position.doc" Stimuli(7, 116) = "Elementary School Position.doc" Stimuli(7, 117) = "FL Lice nse Requirements.html" Stimuli(7, 118) = "Real Estate Courses List.html" Stimuli(7, 119) = "Salary Info.doc" Stimuli(7, 120) = "Interviewer Contact Info.doc" Stimuli(7, 121) = "Intern Benefits.html" Stimuli(7, 122) = "Start End Dates.doc" Stimuli(7, 123) = "Receptionist Intern Posting.html" Stimuli(7, 124) = "Psychometrician Intern Posting.html" Stimuli(7, 125) = "Possible Weekly Schedules.doc" Stimuli(7, 126) = "Background Reading.doc" Stimuli(7, 127) = "Volunteers List.txt" Stimuli(7, 128) = "Supervisor Contact Info.html"

PAGE 124

124 APPENDIX B EXPERIMENT 2: TREE B FOLDER/FILE NAMES Stimuli(0, 1) = "Lecture Notes" Stimuli(1, 1) = "Problem Sets" Stimuli(1, 2) = "Term Paper" Stimuli(2, 1) = "Course Info" Stimuli(2, 2) = "Exam Reviews" Stimuli(2, 3) = "Old Exams" Stimuli(2, 4) = "Course Projects" Stimuli(3, 1) = "School Expenses" Stimuli(3, 2) = "Course Related C osts" Stimuli(3, 3) = "Tuition and Fees" Stimuli(3, 4) = "Books and Supplies" Stimuli(3, 5) = "Lecture Videos" Stimuli(3, 6) = "Articles" Stimuli(3, 7) = "GRE Prep Course" Stimuli(3, 8) = "UF Honor Societies" Stimuli(4, 1) = "Cognitive Psychology" Stimuli(4, 2) = "Expected Credits" Stimuli(4, 3) = "General Anthropology" Stimuli(4, 4) = "Transportation Expenses" Stimuli(4, 5) = "On Campus Meal Plans" Stimuli(4, 6) = "Required Psychology Major Courses" Stimuli(4, 7) = "Gas Expenses" Stimuli(4, 8) = "Grad School Applications" Stimuli(4, 9) = "Music Videos" Stimuli(4, 10) = "Add Drop Period" Stimuli(4, 11) = "Online Journal Entr ies" Stimuli(4, 12) = "Junior Year" Stimuli(4, 13) = "Transcripts" Stimuli(4, 14) = "On Campus Work Study Info" Stimuli(4, 15) = "Basketball Intramural Team" Stimuli(4, 16) = "UF Rugby Team" Stimuli(5, 1) = Required Biology Courses" Stimuli(5, 2) = "Behavioral Analysis" Stimuli(5, 3) = "Previous Balances" Stimuli(5, 4) = "Rental Costs" Stimuli(5, 5) = "Formatting Requirements" Stimuli(5, 6) = "Literature (Poems and Shor t Stories)" Stimuli(5, 7) = "MasterCard" Stimuli(5, 8) = "Bright Futures" Stimuli(5, 9) = "Infirmary" Stimuli(5, 10) = "On Campus Parking" Stimuli(5, 11) = "Courses for Fun" Stimuli(5, 12) = "Publishers" Stimuli(5, 13) = "Water Bill"

PAGE 125

125 Stimuli(5, 14) = "Electric (GRU) Bill" Stimuli(5, 15) = "Prospective Jobs" Stimuli(5, 16) = "SAT Scores" Stimuli(5, 17) = "Restaurants" Stimuli(5, 18) = "Pictures" Stimuli(5, 19) = "Academic Calendars" Stimuli(5, 20) = "Job Interviews" Stimuli(5, 21) = "Confidential Data" Stimuli(5, 22) = "High School Diaries" Stimuli(5, 23) = "Commencement Ceremony Info" Stimuli(5, 24) = "Class Reunion" Stimuli(5, 25) = "Potential Supervisors" Stimuli(5, 26) = "Letters of Recommendation" Stimuli(5, 27) = "Psychology Job Applications" Stimuli(5, 28) = "Tax Returns" Stimuli(5, 29) = "Aerobics Course" Stimuli( 5, 30) = "Dance Club" Stimuli(5, 31) = "Intramural Sports" Stimuli(5, 32) = "Entertainment Expenses" Stimuli(6, 1) = "Statistics" Stimuli(6, 2) = "Pre Med Courses" Stimuli(6, 3) = "Psychology Club" Stimuli(6, 4) = "Health Care Providers" Stimuli(6, 5) = "Potential Investments" Stimuli(6, 6) = "Credit Line Increase" Stimuli(6, 7) = "Merchandise Credits" Stimuli(6, 8) = "Online Transactions" Stimuli(6, 9) = "Cover Letters" Stimuli(6, 10) = "Important Emails" Stimuli(6, 11) = "American History Notes" Stimuli(6, 12) = "Letters from Family" Stimuli(6, 13) = "Discover Card" Stimuli(6, 14) = "Current Online Bets" Stimuli(6, 15) = Scholarship Funds" Stimuli(6, 16) = "Student Loans" Stimuli(6, 17) = "Shands Hospital" Stimuli(6, 18) = "Mental Hospital" Stimuli(6, 19) = "Travel Expenditures" Stimuli(6, 20) = "UF Financial Aid" Stimuli(6, 21) = "Required General Education Classes" Stimuli(6, 22) = "Theme Parks" Stimuli(6, 23) = "Take Home Tests" Stimuli(6, 24) = "Human Resources" Stimuli(6, 25) = "Cruises" Stimuli(6, 26) = "Health Insurance" S timuli(6, 27) = "Phone Bills"

PAGE 126

126 Stimuli(6, 28) = "Car Insurance" Stimuli(6, 29) = "Bank Statements" Stimuli(6, 30) = "Expense Account Balance" Stimuli(6, 31) = "Accounting Info" Stimuli(6, 32) = "Resume" Stimul i(6, 33) = "Presentations" Stimuli(6, 34) = "Hotels" Stimuli(6, 35) = "Songs" Stimuli(6, 36) = "Potential Residence Locations" Stimuli(6, 37) = "Quarterly Sales" Stimuli(6, 38) = "Tutor Contact Information" S timuli(6, 39) = "List of Suppliers" Stimuli(6, 40) = "Marketing Strategies" Stimuli(6, 41) = "Weekly Schedule" Stimuli(6, 42) = "Potential Advisors" Stimuli(6, 43) = "Meeting Times" Stimuli(6, 44) = "Old Love Letters Stimuli(6, 45) = "Senior Year" Stimuli(6, 46) = "Workers' Compensation" Stimuli(6, 47) = "Distinguished Speakers" Stimuli(6, 48) = "High School Football" Stimuli(6, 49) = "Corporate Structure" Stimuli(6, 50 ) = "Signing Bonus" Stimuli(6, 51) = "Volunteer Opportunities" Stimuli(6, 52) = "Research Assistant" Stimuli(6, 53) = "Non Psychology Job Applications" Stimuli(6, 54) = "Work Schedule" Stimuli(6, 55) = "Organize Thes e Data" Stimuli(6, 56) = "Potential Business Partners" Stimuli(6, 57) = "Gym Membership" Stimuli(6, 58) = "Approved Dietary Supplements" Stimuli(6, 59) = "Marching Band" Stimuli(6, 60) = "Social Calendar" Sti muli(6, 61) = "Drug Prescriptions" Stimuli(6, 62) = "Gator Games Pictures" Stimuli(6, 63) = "Christmas Wish List" Stimuli(6, 64) = "Oscar Winning Movies" Stimuli(7, 1) = "Signup.html" Stimuli(7, 2) = "Cancellation.ht ml" Stimuli(7, 3) = "July '06 Consumption.doc" Stimuli(7, 4) = "August '06 Consumption. doc" Stimuli(7, 5) = "December '06 Statement.pdf" Stimuli(7, 6) = "January '07 Statement.pdf" Stimuli(7, 7) = "October '06 State ment.pdf" Stimuli(7, 8) = "November '06 Statement.pdf" Stimuli(7, 9) = "Laptop Return Receipt.jpg"

PAGE 127

127 Stimuli(7, 10) = "Watch Return Receipt.jpg" Stimuli(7, 11) = "Payment History.doc" Stimuli(7, 12) = "September '06 St atement.pdf" Stimuli(7, 13) = "Olive Garden Redemption.html" Stimuli(7, 14) = "Foot Locker Redemption.html" Stimuli(7, 15) = "General Info.doc" Stimuli(7, 16) = "Petition.doc" Stimuli(7, 17) = "3rd Year Tuition.doc" Stimuli(7, 18) = "3rd Year Fees.doc" Stimuli(7, 19) = "4th Year Tuition.doc" Stimuli(7, 20) = "4th Year Fees.doc" Stimuli(7, 21) = "On Campus Bookstore Purchase.bmp" Stimuli(7, 22) = "Off Campus Bookstore Purchase.bm p" Stimuli(7, 23) = "Packet Costs.doc" Stimuli(7, 24) = "Software Costs.doc" Stimuli(7, 25) = "Citation Total.doc" Stimuli(7, 26) = "Payment Receipt.jpg" Stimuli(7, 27) = "Fall '06.txt" Stimuli(7, 28) = "Spri ng '07.txt" Stimuli(7, 29) = "Football Season Ticket Receipt.jpg" Stimuli(7, 30) = "Performance Show Ticket Prices.doc" Stimuli(7, 31) = "Declining Balance Account Summary.html" Stimuli(7, 32) = "Vending Account Balance.html Stimuli(7, 33) = "Christmas '06.jpg" Stimuli(7, 34) = "Thanksgiving '06.jpg" Stimuli(7, 35) = "Europe '05.bmp" Stimuli(7, 36) = "California '04.bmp" Stimuli(7, 37) = "Football '06.jpg" Stimuli(7, 38) = "Bas ketball '06.jpg" Stimuli(7, 39) = "Key West '05.jpg" Stimuli(7, 40) = "Bahamas '04.jpg" Stimuli(7, 41) = "Shakira Hips Don't Lie.mp3" Stimuli(7, 42) = "Britney Spears Everytime Lyrics.html" Stimuli(7, 43) = "50 C ent Candy Shop.mp3" Stimuli(7, 44) = "Snoop Dog Drop It Like It's Hot.mp3" Stimuli(7, 45) = "LeAnn Rimes Blue.mp3" Stimuli(7, 46) = "Keith Urban You'll Think Of Me.mp3" Stimuli(7, 47) = "Green Day Time Of Your Life .mp3" Stimuli(7, 48) = "Beatles I Want To Hold Your Hand.mp3" Stimuli(7, 49) = "To Ex.doc" Stimuli(7, 50) = "From Ex.doc" Stimuli(7, 51) = "From Mom.doc" Stimuli(7, 52) = "To Dad.doc" Stimuli(7, 53) = "Fres hman Year Entries.txt" Stimuli(7, 54) = "Sophomore Year Entries.txt" Stimuli(7, 55) = "Entries Only Friends Can See.html"

PAGE 128

128 Stimuli(7, 56) = "Entries Only I Can See.html" Stimuli(7, 57) = "Robert Frost Stars.pdf" Stimuli(7, 58) = "Maya Angelou Still I Rise.pdf" Stimuli(7, 59) = "James Joyce Clay.pdf" Stimuli(7, 60) = "Nikolai Gogol The Cloak.pdf" Stimuli(7, 61) = "How To Be A Leader.html" Stimuli(7, 62) = "Eating Healthy.doc" Stimuli(7, 63) = "Blueberry Crepes.html" Stimuli(7, 64) = "Pumpkin Pie.html" Stimuli(7, 65) = "Working Memory.ppt" Stimuli(7, 66) = "Problem Solving.ppt" Stimuli(7, 67) = "Exam 2 Review.doc" Stimuli(7, 68) = "Final Exam Review.doc" Stimuli(7, 69) = "Abstract.doc" Stimuli(7, 70) = "References.doc" Stimuli(7, 71) = "Syllabus.doc" Stimuli(7, 72) = "Blog Comments.html" Stimuli(7, 73) = "Fall 2006 Exam 1.doc" Stimuli( 7, 74) = "Spring 2005 Exam 2.doc" Stimuli(7, 75) = "Skeletal Biological Distances.pdf" Stimuli(7, 76) = "Man: The Social Animal.pdf" Stimuli(7, 77) = "Normal Distribution.doc" Stimuli(7, 78) = "Probability.doc" Stimu li(7, 79) = "Data Collection.xls" Stimuli(7, 80) = "Data Analysis.xls" Stimuli(7, 81) = "Psychology Club Members.txt" Stimuli(7, 82) = "Future Therapists Members.txt" Stimuli(7, 83) = "Civil Engineers Society Info.txt" Stimuli(7, 84) = "Architecture Club Info.txt" Stimuli(7, 85) = "Golden Key Requirements.html" Stimuli(7, 86) = "Phi Beta Kappa Requirements.html" Stimuli(7, 87) = "Pi Gamma Mu Regional Members.xls" Stimuli(7, 88) = "Psi C hi Requirements.html" Stimuli(7, 89) = "Members List.doc" Stimuli(7, 90) = "Rehearsal Schedule.doc" Stimuli(7, 91) = "Practice Schedule.pdf" Stimuli(7, 92) = "Contacts.txt" Stimuli(7, 93) = "Roster.doc" Stimu li(7, 94) = "Game Schedule.pdf" Stimuli(7, 95) = "Rules and Regulations.html" Stimuli(7, 96) = "Playbook.ppt" Stimuli(7, 97) = "Step By Step Instructions.pdf" Stimuli(7, 98) = "Interactive Instructions.html" Stimuli( 7, 99) = "History Of Psychology Outline.doc" Stimuli(7, 100) = "Famous Psychologists Outline.doc" Stimuli(7, 101) = "Quantitative Materials.ppt"

PAGE 129

129 Stimuli(7, 102) = "Verbal Materials.ppt" Stimuli(7, 103) = "Confirmation #.html Stimuli(7, 104) = "Test Location.html" Stimuli(7, 105) = "Dr. White's Letter.doc" Stimuli(7, 106) = "Dr. Fischler's Letter.doc" Stimuli(7, 107) = "Community College Transcript.pdf" Stimuli(7, 108) = "University of Florida Transcript.pdf" Stimuli(7, 109) = "Clinical Psych Program App.pdf" Stimuli(7, 110) = "Industrial Org Psych Program App.pdf" Stimuli(7, 111) = "Cognitive Psych Program App.pdf" Stimuli(7, 112) = "Behavioral Analysis P rogram App.pdf" Stimuli(7, 113) = "Participant Schedule.html" Stimuli(7, 114) = "Lab Contact Info.txt" Stimuli(7, 115) = "Middle School Position.doc" Stimuli(7, 116) = "Elementary School Position.doc" Stimuli(7, 117) = "FL License Requirements.html" Stimuli(7, 118) = "Real Estate Courses List.html" Stimuli(7, 119) = "Salary Info.doc" Stimuli(7, 120) = "Interviewer Contact Info.doc" Stimuli(7, 121) = "Intern Benefits.html" Stimul i(7, 122) = "Start End Dates.doc" Stimuli(7, 123) = "Receptionist Intern Posting.html" Stimuli(7, 124) = "Psychometrician Intern Posting.html" Stimuli(7, 125) = "Possible Weekly Schedules.doc" Stimuli(7, 126) = "Background R eading.doc" Stimuli(7, 127) = "Volunteers List.txt" Stimuli(7, 128) = "Supervisor Contact Info.html"

PAGE 130

130 LIST OF REFERENCES Altmann, E.M. & John, B.E. (1999). Episodic indexing: A model of memory for attention events. Cognitive Science, 23 11 7 156. Anderson, J.R., Matessa, M., & Lebiere, C. (1997). ACT R: A theory of higher levelcognition and its relation to visual attention. Human Computer Interaction, 12, 439 462. Atwood, M.E.& Polson, P.G. (1976). A process model for water jug problems. Cogn itive Psychology, 8, 191 216. Benyon, D. & Hook, K. (1997). Navigation in information spaces: Supporting the individual. Human 39 46. Boardman, R & Sasse, M.A., ss tool ACM Conference on Human Factors in Computing Systems, CHI Letters 6(1), 2004 Campbell, D.J. (1988). Task complexity: A review and analysis. Academy of Management Review, 13, 40 52. Card, S.K. (1 984). Visual search of computer command menus. In H. Bouma & D.G. Bouwhuis (Eds.) Attention and Performance X, Control of Language Processes Hillsdale, NJ: Lawrence Erlbaum Associates. Card, S.K., Moran, T.P., & Newell, A. (1983). The psychology of human computer interaction Hillsdale, NJ: Lawrence Erlbaum Associates. Davies, S.P. (2000). Memory and planning processes in solutions to well structured problems. The Quarterly Journal of Experimental Psychology, 53, 896 927. Delaney, P.F., Ericsson, K.A., & Knowles, M.E. (2004). Immediate and sustained effects of planning in a problem solving task. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30 1 15. Delaney, P.F., Reder, L.M., Staszewski, J.J., & Ritter, F.E. (1998). The strategy s pecific nature of improvement: The power law applies by strategy within task. Psychological Science, 9, 1 7. Dix A. & Marshall J (2003). At the right time: when to sort web history and bookmarks. Proceedings of Human Computer Interaction International 2 003 Lawrence Erlbaum Associates, 758 762 Dumais, S.T. & Jones, W.P. (1985) A comparison of symbolic and spatial fili ng. In Proceedings of the SIGCHI conference on Human factors in computing systems ACM Press, 127 130. Dunbar, K. (1998). Problem solving In W. Bechtel & G. Graham (Eds.), The MIT encyclopedia of the cognitive sciences (pp. 448 451). Cambridge, MA: MIT Press.

PAGE 131

131 Ericsson, K.A. & Simon, H.A. (1993). Protocol analysis; Verbal reports as data (revised edition ) Cambridge, MA: Bradford books/MIT Press. Franzke, M (1995). Turning research into practice: characteristics of display based interaction. Proceedings of the Conference on Human Factors in Computing Systems, New York, Asssociation for Computing Machinery, 421 428. Fu, W. T., & Gray, W. D ( 2006). Suboptimal tradeoffs in information seeking. Cognitive Psychology, 52 (3), 195 242. Garner, W.R. (1974). The processing of information and structure New York: Wiley. Glanzer, M. & Cunitz, A.R. (1966). Two storage mechanisms in free recall. Journa l of Verbal Learning and Verbal Behaviour, 5, 351 360. Gunzelmann, G. & Blessing, S. B. (2000). Why are some problems easy? New insights into the Tower of Hanoi. In L. R. Gleitman and A. K. Joshi (Eds.), Proceedings of the Twenty Second Annual Conference o f the Cognitive Science Society (p. 1029). Mahwah, NJ: Lawrence Erlbaum Associates. Hitch, G. & Baddeley, A.D. (1976). Verbal reasoning and working memory. The Quarterly Journal of Experimental Psychology, 28, 603 621. Howes, A. (1994). A model of the acqu isition of menu knowledge by exploration. In B. Adelson, S. Dumais, J. Olson (Eds.) (pp. 445 451), Boston, MA: ACM Press. Howes, A. & Payne, S.J. (2001). The strategic use of memory for frequency and recency in search control. 23 rd Annual Conference of the Cognitive Science Society Edinburgh, Scotland. Johnson, E.J., Bellman, S., Lohse, G.L. (2003). Cognitive lock in and the power law of practice. Journal of Marketing, 67, 62 75. Kennedy, R.S. & Stan ney, K.M. (1998). Aftereffects from virtual environment exposure: How long do they last?. Proceedings of the 42 nd Annual Meeting of the Human Factors and Ergonomics Society 1476 1480. Kotovsky, K., Hayes, J.R., & Simon, H.A. (1985). Why are some problems hard? Evidence from Tower of Hanoi. Cognitive Psychology, 17, 248 294. Kotovsky, K. & Simon, H.A. (1990). What makes some problems really hard: Explorations in the problems space of difficulty. Cognitive Psychology, 22, 143 183. Reprinted in Simon, H.A. (1989). Models of Thought, Volume Two. Yale University Press, New Haven, CT. Lachman, R., Lachman, J.L., & Butterfield, E.C. (1979). Cognitive psychology and information processing: An introduction Hillsdale, NJ: Lawrence Erlbaum Associates.

PAGE 132

132 Landauer, T.K ., Foltz, P.W., & Laham, D. (1998). Introduction to latent semantic analysis. Discourse Processes, 25, 259 284. Lefebvre, H. (1991). The production of space Oxford: Blackwell. Lockhart, R.S. (2002). Levels of processing, transfer appropriate processing, a nd the concept of robust encoding. Memory, 10 397 403. Logan, G.D. (1988). Toward an instance theory of automatization. Psychological Review, 95, 492 527. Lovett, M.C. (2002). Problem solving. In D. Medin (Ed.), psycholog y (pp. 317 362). New York: Wiley. Lovett, M.C. & Anderson, J.R. (1996). History of success and current context in problem solving: Combined influences on operator selection. Cognitive Psychology, 31, 168 217. Lynch, G., Palmiter, S., & Tilt, C. (1999). Th e Max model: A standard web site user model. Proceedings of the Fifth Conference on Human Factors and the Web. Gaithersburg, MD. Maccoby, E.E. & Jacklin, C.N. (1974). The psychology of sex differences Stanford, CA: Stanford University Press. Malone, T.W. (1983). How do people organize their desks? Implications for the design of office information systems. ACM Transactions on Office Information Systems, 1, 99 112. McCauley, M.E. (1992). Cybersickness: Perception of self motion in virtual environments. Pres ence: Teleoperators and Virtual Environments, 1, 311 318. Medin, D., Goldstone, R., & Genter, D. (1993). Respects for similarity. Psychological Review, 100, 254 278. Newell, A., Shaw, J.C., & Simon, H.A. (1957). Empirical explorations of the logic theory m achine. Proceedings of the Western Joint Computer Conference 218 239. Newell, A. & Simon, H.A. (1972). Human Problem Solving Englewood Cliffs, NJ: Prentice Hall. Newell, B.R., & Shanks, D.R. (2004). On the role of recognition in decision making. Journ al of Experimental Psychology: Learning, Memory, and Cognition, 30, 923 935. Nosofsky, R.M. & Palmeri, T.J. (1997). Comparing exemplar retrieval and decision bound models of speeded perceptual classification. Perception & Psychophysics, 59, 1027 1048. ra, K.P. & Payne, S.J. (1999). Planning and the user interface: the effects of lockout time and error recovery cost. International Journal of Human Computer Studies,50, 41 59. n planfulness of problem solving and learning. Cognitive Psychology, 35, 34 70.

PAGE 133

133 Payne, S.J., Richardson, J., & Howes, A. (2000). Strategic use of familiarity in display based problem solving. Journal of Experimental Psychology Learning, Memory, and Cogni tion, 2, 1685 1701. Proctor, R.W. & Vu, K.L. (2003). Human information processing: An overview for human computer interaction. In J.A. Jacko & A. Sears (eds.) The human computer interaction handbook: Fundamentals, evolving technologies and emerging applica tions Mahwah, NJ: Lawrence Erlbaum Associates. Raab, D.H. (1962). Statistical facilitation of simple reaction times. Transactions of the New York Academy of Sciences, 24, 574 590. Reder, L.M. (1982). Plausibility judgments vs. fact retrieval: Alternative strategies for sentence verification. Psychological Review, 89, 250 280. Reder, L.M. & Ritter, F.E. (1992). What determines initial feeling of knowing? Familiarity with question terms, not with the answer. Journal of Experimental Psychology: Learning, Mem ory, and Cognition, 18, 435 451. Reed, S.K., Ernst, G.W., & Banerji, R. (1974). The role of analogy in transfer between similar problem states. Cognitive Psychology, 6 436 450. Richardson, J., Howes, A., & Payne, S.J. (1997). An empirical investigation of memory for routes through menu structures. Human Sydney, Australia. Rieman, J. (1994). Learning strategies and exploratory behavior of interactive computer users Ph.D. Thesis, Department of Computer Science, University of Colorado, Boulder, CO. Rieman, J., Young, R.M., & Howes, A. (1996). A dual space model of iteratively deepening exploratory learning. International Journal of Human Computer Studies, 44, 743 745. Schunn, C.D., Reder, L.M., Nhouyvanisvong, A., Richards, D.R., Stroffolino, P.J. (1997). To role in strategy selection. Journal of Experimental Psychology: Learning Memory, and Cognition, 23, 3 29. Shields, R. (1991). Places on the Margin: Alternative Geographies of Modernity .Routledge. Stanney, K.M., Mourant, R.R., & Kennedy, R.S. (1998). Human factors issues in virtual environments: A review of the literature. Presence, 7 (4), 327 351. Tolman, E.C., Ritchie, B.F., & Ka lish, D. (1946). Studies in spatial learning: II. Place learning versus response learning. Journal of Experimental Psychology, 36 221 229. Townsend, J.T. & Ashby, F.G. (1983). Stochastic modeling of elementary psychological processes. New York: Cambridge University.

PAGE 134

134 Trudel, C. I. & Payne, S.J. (1995). Reflection and goal management in exploratory learning. International Journal of Human Computer Studies, 42, 307 339. Tversky, A. & Kahneman, D. (1973). Availability: A heuristic for judging frequency and pr obability. Cognitive Psychology, 4, 207 232. Walker, J. (1990). Through the looking glass. In B. Laurel (ed.) The Art of Human Computer Interface Design Cambridge, MA: Addison Wesley. Waltz, J.A., Lau, A., Grewal, S.K., & Holyoak, K.J. (2000). The role o f working memory inanalogical mapping. Memory & Cognition, 28, 1205 1212. Whitt aker, S. Terveen, L., & Nardi, B. A. (2000) start addressing it: A reference task agenda for HCI. Human Computer Interaction 15 75 106. Wi lson, J.R., Nichols, S., Haldane, C. (2000). Measurement of presence and its consequences in virtual environments. International Journal of Human Computer Studies, 52, 471 491. Zhang, J. (1997). The nature of external representations in problem solving. Co gnitive Science, 21, 179 217. Zhang, J. & Norman, D.A. (1994). Representations in distributed cognitive tasks. Cognitive Science, 18, 87 122.

PAGE 135

135 BIOGRAPHICAL SKETCH The author received a B.S. in behavioral neuroscience from Trinity College in 1998. He then completed two graduate degrees at the University of South Florida, an M.A. in psychology (2001) and an M.S. in management information systems (2003). Keith received a Ph.D. in psychology, focusing on the cognitive psychology of human computer interaction, from the University of Florida in 2007. Having completed software usability engineering internships at Microsoft Corp. and Google Inc., Keith is currently employed as a Usability Analyst at the Microsoft account of Siemens IT Solutions and Services. Pre sently, he is part of the hardware group at Microsoft, conducting user research on PC mice and other interactive gaming devices.