<%BANNER%>

Marketing Indoor Plants as Air Cleaners

Permanent Link: http://ufdc.ufl.edu/UFE0043708/00001

Material Information

Title: Marketing Indoor Plants as Air Cleaners a Choice-Based Conjoint Analysis
Physical Description: 1 online resource (111 p.)
Language: english
Creator: Solano, Alexis A
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2012

Subjects

Subjects / Keywords: conjoint -- experiment -- floriculture -- logit -- marketing -- surveys
Food and Resource Economics -- Dissertations, Academic -- UF
Genre: Food and Resource Economics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Florida's floriculture industry had experienced a decrease in sales beginning in 2007. This decrease coincided with the recession of 2007. In 2009 there was some recovery but sales were not up to their 2006 level. To improve these sales the industry is search for a new way to market indoor plants. One possible avenue is to market indoor plants as "green" or natural indoor air cleaners. Scientific research has shown that specific indoor plants can remove indoor air pollution, sometimes called Volatile Organic Compounds/Chemicals (VOCs). In order to determine if consumers are interested in buying these indoor plants choice-based conjoint (CBC) analysis was used. CBC analysis is a commonly used marketing tool and is most often in a survey form as was done here. This method allows participants to choose the attribute levels that they would prefer to have in a houseplant. Usually the attributes are chosen by the researcher and/or a marketing manager. In this study, however, the majority of participants were allowed to select the attributes they preferred. The survey permitted participants to choose three of six attributes. In addition, information about VOCs and the ability of specific indoor plants to remove them was randomly provided to some participants. This information was provided to determine if it had an effect on participants' choices. The results of the CBC analysis when VOC information was provided were compared to the results of the CBC analysis when VOC information was not given. To determine if there was a difference between surveys allowing participants to choose attributes and a survey with a set number of attributes, a survey with a fixed number of attributes was also distributed. Again, VOC information was randomly provided to participants. The results show that VOC information did make a difference in participants' selections. There were also small differences between the surveys permitting attribute selection and the surveys with fixed attributes.
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 Alexis A Solano.
Thesis: Thesis (Ph.D.)--University of Florida, 2012.
Local: Adviser: House, Lisa O.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2012
System ID: UFE0043708:00001

Permanent Link: http://ufdc.ufl.edu/UFE0043708/00001

Material Information

Title: Marketing Indoor Plants as Air Cleaners a Choice-Based Conjoint Analysis
Physical Description: 1 online resource (111 p.)
Language: english
Creator: Solano, Alexis A
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2012

Subjects

Subjects / Keywords: conjoint -- experiment -- floriculture -- logit -- marketing -- surveys
Food and Resource Economics -- Dissertations, Academic -- UF
Genre: Food and Resource Economics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Florida's floriculture industry had experienced a decrease in sales beginning in 2007. This decrease coincided with the recession of 2007. In 2009 there was some recovery but sales were not up to their 2006 level. To improve these sales the industry is search for a new way to market indoor plants. One possible avenue is to market indoor plants as "green" or natural indoor air cleaners. Scientific research has shown that specific indoor plants can remove indoor air pollution, sometimes called Volatile Organic Compounds/Chemicals (VOCs). In order to determine if consumers are interested in buying these indoor plants choice-based conjoint (CBC) analysis was used. CBC analysis is a commonly used marketing tool and is most often in a survey form as was done here. This method allows participants to choose the attribute levels that they would prefer to have in a houseplant. Usually the attributes are chosen by the researcher and/or a marketing manager. In this study, however, the majority of participants were allowed to select the attributes they preferred. The survey permitted participants to choose three of six attributes. In addition, information about VOCs and the ability of specific indoor plants to remove them was randomly provided to some participants. This information was provided to determine if it had an effect on participants' choices. The results of the CBC analysis when VOC information was provided were compared to the results of the CBC analysis when VOC information was not given. To determine if there was a difference between surveys allowing participants to choose attributes and a survey with a set number of attributes, a survey with a fixed number of attributes was also distributed. Again, VOC information was randomly provided to participants. The results show that VOC information did make a difference in participants' selections. There were also small differences between the surveys permitting attribute selection and the surveys with fixed attributes.
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 Alexis A Solano.
Thesis: Thesis (Ph.D.)--University of Florida, 2012.
Local: Adviser: House, Lisa O.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2012
System ID: UFE0043708:00001


This item has the following downloads:


Full Text

PAGE 1

1 MARKETING INDOOR PLANTS AS AIR CLEANERS: A CHOICE BASED CONJOINT ANALYSIS By ALEXIS ARMSTRONG SOLANO 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 2012

PAGE 2

2 2012 Alexis Armstrong Solano

PAGE 3

3 To my mother, father, and sister who supported me through everything, to Ledia for being there whenever I needed her, to Gabi, Trent, Jess, and Francesca for being great friends, to Bethany who always pushed me to take on more challenges

PAGE 4

4 ACKNOWLEDGEMENTS I thank my chair, Dr. Lisa House, my cochair, Dr. Edward Evans, and my committee members, Dr. Zhifeng Gao and Dr. Tracy Irani for their guidance and support. I thank the National Foliage Foundation for its generous support. I thank my parents and my sister for being there when I needed them the most.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ...............................................................................................................4 LIST OF TABLES ...........................................................................................................................7 LIST OF FIGURES .........................................................................................................................9 ABSTRACT ...................................................................................................................................10 CHAPTER 1 INTRODUCTION ..................................................................................................................12 Floridas Floriculture Industry from 2000 to 2010 ................................................................12 Floridas Housing Market ......................................................................................................14 Importance of Indoor Plants ...................................................................................................16 Objectives ...............................................................................................................................18 First Objective: Consumer Preferences ..........................................................................19 Second Objective: Theoretical Implications of This Study ............................................19 2 LITERATURE REVIEW .......................................................................................................23 Studies on VOC Removal ...................................................................................................... 23 Consumer Preferences, Trends, and Marketing Strategies in the Floriculture Industr y .........28 3 THEORETICAL FRAMEWORK .........................................................................................33 Guidelines and Principles of Conjoint Analysis ....................................................................33 Theoretical Basis of Conjoint Analysis ..................................................................................36 Selecting Attributes ...............................................................................................................38 Data Collection ......................................................................................................................40 Data Collection Method: the Full Profile Method ..........................................................41 Data Collection Method: the Tradeoff Matrix Approach ...............................................42 Data Collection Method: the Method of Paired Comparisons ........................................42 Data Collection Method: Adaptive Conjoint Analysis ...................................................43 Data Collection Method: Choice Based Conjoint Analysis ...........................................43 Data Collection Method: Two Other Approaches ..........................................................44 Fractional Factorial Conjoint and ChoiceBased Conjoint Analysis: a Comparison ............45 Evaluation of the Attributes ...................................................................................................47 Selection of the Data Collection Procedure ...........................................................................47 Data Analysis .........................................................................................................................48 Data Analysis: Model Specification ...............................................................................48 Within attribute constraints ........................................................................................49 Across attributes const raints .......................................................................................51

PAGE 6

6 Zero one across attribute constraints ..........................................................................52 Ordinal inequality across attribute constraints ...........................................................54 Across subject constraints ..........................................................................................55 Estimation Method .................................................................................................................56 Evaluation and Selection of the Model ............................................................................57 Market Simulations ................................................................................................................58 Tests of Reliability .................................................................................................................59 4 METHODOLOGY .................................................................................................................62 Construction of the Surveys ...................................................................................................62 Preliminary Results ................................................................................................................66 5 RESULTS ...............................................................................................................................73 Attributes Selected by Participants .........................................................................................73 Regression Analysis ................................................................................................................73 Conditional Logit Results .......................................................................................................75 Surveys with Attribute Selection .....................................................................................78 Surveys with Fixed Attributes .........................................................................................78 Discussion and Comparison of Willingness to Pay for Attributes ..................................79 VOC removal .............................................................................................................79 Flowering and tags .....................................................................................................79 Toxicity .......................................................................................................................80 Sunlight .....................................................................................................................81 Hardiness ....................................................................................................................81 Height ........................................................................................................................82 Fixed attributes ...........................................................................................................83 Attribute selection and fixed attributes ......................................................................83 Ranges and Weighted Averages of Willingness to Pay for Each Attribute ..........................84 6 CONCLUSION AND FUTURE RESEARCH .....................................................................98 REFERENCE LIST .....................................................................................................................107 BIOGRAPHICAL SKETCH .......................................................................................................111

PAGE 7

7 LIST OF TABLES Table page 41 Attributes and their corresponding attributes .....................................................................69 42 Frequency of purchasing indoor plants (attribute selection) ..............................................69 43 Place of purchase (can purchase at multiple stores) ...........................................................69 44 Problems finding plant variety. ..........................................................................................69 45 Gender of participants ........................................................................................................70 46 Ethnicity of participants (select all that apply) ..................................................................70 47 Ages of participants ...........................................................................................................70 48 Number of children living in the household (select all that apply) ....................................70 49 Pets in living in the household (select all that apply) ........................................................70 410 Ownership of home ............................................................................................................71 411 Type of home .....................................................................................................................71 412 Primary shopper in the household......................................................................................71 413 Marital status of participant ...............................................................................................71 414 Respiratory problems .........................................................................................................71 415 Allergies (select all that apply) ..........................................................................................72 416 Annual household income ..................................................................................................72 51 Number of participants per survey attribute selection .....................................................86 52 Parameter estimates and WTP for flexible attribute models .............................................87 53 Significance and sign of attribute levels ............................................................................92 54 Parameter estimates and WTP for the fixed attribute model .............................................95 55 Willingness to pay for each attribute ................................................................................96 56 Willingness to pay weighted means ..................................................................................97

PAGE 8

8 61 Order of preferences for attributes (descending) ............................................................106

PAGE 9

9 LIST OF FIGURES Figure page 11 Total foliage sales by producer from 2000 to 2010. ..........................................................20 12 Foliage sales by type from 2000 to 2010 ...........................................................................20 13 Wholesale value of sales from 2000 to 2010 .....................................................................21 14 Number of producers from 2000 to 2010 ..........................................................................21 15 Existing home and condominium sales from 2006 to 2010 ...............................................22 16 Median home prices from 2006 to 2010 ............................................................................22

PAGE 10

10 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 MARKETING INDOOR PLANTS AS AIR CLEANERS: A CHOICE BASED CONJOINT ANALYSIS By Alexis Armstrong Solano May 2012 Chair: Lisa House Major: Food and Resource Economics Florida s floriculture industry had experienced a decrease in sales beginning in 2007. This decrease coincided with the recession of 2007. In 2009 and 2010 there was some recovery but sales were not up to their 2006 level. To improve these sales the industry i s search for a new way to market indoor plants. One possible avenue is to market indoor plants as green or natural indoor air cleaners. Scientific research has shown that specific indoor plants can remove indoor air pollution, sometimes called Volatile Organic Compounds/Chemicals (VOCs). In order to determine if consumers are interested in buying these indoor plants choice based conjoint (CBC) analysis was used. CBC analysis is a commonly used marketing tool and is most often in a survey form as was done here. This method allows participants to choose the attribute levels that they would prefer to have in a houseplant. Usually the attributes are chosen by the researcher and/or a marketing manager. In this study, however, the majority of participants were allowed to select the attributes they preferred. The survey permitted participants to choose three of six attributes. In addition, information about VOCs and the ability of specific indoor plants to remove them was randomly provided to some participants. This information was provided to

PAGE 11

11 determine if it had an effect on participants choices. The results of the CBC analysis when VOC information was provided were compared to the results of the CBC analysis when VOC information was not given. To dete rmine if there was a difference between surveys allowing participants to choose attributes and a survey with a set number of attributes, a survey with a fixed number of attributes was also distributed. Again, VOC information was randomly provided to parti cipants. The results show that VOC information did make a difference in participants selections. There were also small differences between the surveys permitting attribute selection and the surveys with fixed attributes.

PAGE 12

12 CHAPTER 1 INTRODUCTION Floridas Floriculture Industry from 2000 to 2010 Figure 1 1 shows floriculture sales from 2000 to 2010. Floridas floriculture sales increased steadily from 2000 (NASS 2002) to 2002 (NASS 2003). These sales decreas ed in 2003 and again in 2004; however, production in 2004 was likely affected by hurricanes (NASS 2005). Sales increased in 2005 but decreased in 2006, again due to hurricanes (NASS 2007). In 2007 floriculture sales increased; however, in late 2007 the r ecession had begun and in 2008 sales were down 5% from the previous year (NASS 2009). As shown in Figure 11 in 2008, floriculture sales from small producers (those with sales of $10,000 or more) were $921.7 million down from $968 million in 2007. In 2009 these sales decreased to $814.9 million and declined even further to $809.6 million in 2010 (NASS 2011). Small producers were not the only ones to experience a decrease in sales. Large p ro ducers (producers with sales of $100,000) also saw a decrease in their sales, from $953 million to $906 million in 2008 (N ASS 2009). These sales decreased to $801.3 million in 2009 and to $788.1 million in 2010 (NASS 2011). Florida is the leader in sales of foliage plants. Figure 12 displays foliage sales and the components of these sales (potted and hanging plants) from 2000 to 2010. Foliage sales for this time period follow the same trend of the floriculture sales. In 2008 the effect of the recession was beginning to show. Large producers had foliage sales of $493.9 million in 2007 but these sales decreased to $480.4 million in 2008 and to $399.8 million in 2009 before increasing in 2010 to $411.8 million (NASS 2011). Floridas producers sell ma ny types of plants, including cut flowers and landscaping plants (Figure 1 3). The sales of these plants have been inconsistent over the last few years.

PAGE 13

13 From 2000 to 2006 these sales exhibit the same pattern as that of floriculture sales. Again in 2008 t hese sales began to decline, with bedding/garden and propagative materials as the exceptions. Cut flower sales decreased from $11.3 million in 2007 to $9 million in 2008 (NASS 2009) and to $8.1 million in 2009 (NASS 2011). Cut flower sales were not avail able for 2010. Cut cultivated greens sales decreased from $74.4 million in 2007 to $70.8 million in 2008 (NASS 2009) but increased to $56 million in 2009 and to $60.7 million in 2010 (NASS 2011) Propagative f loriculture material sales decreased from $1 11.8 million in 2007 to $77.4 million in 2008 (NASS 2009). Sales for these materials then increased in 2009 to $85.8 million before decreasing to $80.1 million in 2010 (NASS 2011). Between 2007 and 2008 s ales for bedding/garden plants did increase (from $104.7 million to $109.9 million ) as did potted flowering plant sales (from $142.8 million to $146.6 million) (NASS 2009) However, sales for bedding/garden plants decreased to $88.4 million and to $73.2 million in 2009 and 2010, respectively. Potted flow ering plant sales also decreased to $130.9 million in 2009 and decreased again to $115.9 million in 2010 (NASS 2011). In addition to the decrease in sales the number of producers has also decreased. Figure 14 shows the changes in the number of producers. From 2000 to 2010 the number of both small and large producers has decreased, though in 2003 there was a large increase in small producers. In 2007, when the recession began, there were 540 large producers compared with 515 in 2008 (N ASS 2009) This decrease continued in 2009 (475 producers) and 2010 (431 producers) (NASS 2011). T he number of small producers increased fr om 869 to 871 from 2007 to 2008 (N ASS 2009) before decreasing in 2009 to 811 producers in 2009 and to 758 producers in 2010 (NASS 2011).

PAGE 14

14 Hurricanes impacted sales for 2004 and 2006; however the decline in sales since 2007 may have had a different cause. Beginning in late 2007 the world economy has been plagued by a recession, which may have affected U.S. floriculture sales from 2008 to the present. In addition the current recession has affected Floridas housing sales. The decrease in home sales may have also played a role in the decline in floriculture sales. Floridas Housing Market According to the U.S. Department of Housing and Urban Development (HUD) the U.S. housing market faced significant declines from 2006 to 2007. Floridas housing market has been especially affected by the recession and faced a severe decline in the housing market during this period. As show n in Figure 15 Florida experienced a 29% decrease in existing single family home sales (from 183,380 homes to 130,200 homes) and a 27% decrease in condominiums sales (from 56,900 homes in to 41,500 homes) (HUD 2007). Median prices (Figure 16) for both e xisting homes and condominiums decreased in this period as well, from $247,100 to $233,600 and $211,500 to $205,100 for existing homes and condominiums, respectively (HUD 2007). This decrease in sales may have contributed to the loss of 18,700 jobs in construction in 2007 (HUD 2007). In 2008 housing sales continued their decline, although this decrease was smaller in magnitude (Figure 15). Florida saw existing home sales decrease by 4% from the previous year, to 124,200 homes and condominium sales dec rease by 9%, to 37,800 condominiums (HUD 2008). The median price of existing homes and condominiums decreased to $187,800 and $164,400, respectively (Figure 1 6) (HUD 2008). Again, construction employment decreased

PAGE 15

15 but the reduction was greater in 2008. Approximately 78,800 construction jobs were lost in 2008 in Florida alone (HUD 2008). There was some improvement in 2009. Figure 15 shows that existing home sales increased, to 163,100 and condominium sale increased to 56,000 (HUD 2009). As shown in F igure 1 6, the median price of an existing home decreased by 24% to $142,600 and the median price for condominiums declined by 34% to $108,000. Data for construction employment were unavailable (HUD 2009). The 2010 housing market showed some more improvement in existing home sales. These sales (Figure 1 5) increased to 170,900, a 5% increase from 2009 and a 38% increase from 2008. The sales for condominiums were 72,050, an increase of 29% from 2009 and a nearly 50% increase from 2008. However, the median price (Figure 1 6) for existing homes decreased by 4% to $136,500 and the median price for condominiums decreased by 15% to $91,300. Again data for construction employment was not available. Sales for both homes and floriculture products have decre ased since 2007. It is highly likely that the decline in home sales and disposable income have adversely impacted the floriculture industry. Indoor Plants require sunlight and room to grow; some consumers may not be able to afford the home that is big enough to allow enough light or room for a houseplant. Therefore, new ways to market indoor plants should be explored. One way that may improve sales is to market indoor plants as natural indoor air cleaners.

PAGE 16

16 Importance of Indoor Plants According to the U.S. Environmental Protection Agency (EPA) the air inside the home can be polluted. This pollution occurs through the spread of dust, dander, pollen, mold spores, smoke particles, and volatile organic compounds (VOCs). VOCs are gasses, such as formaldehy de, that are produced by household products like paint strippers, pesticides, air fresheners, and glues. There are thousands of sources of VOCs ( Indoor Air Quality 2010a). The levels of organic pollutants inside most homes were found to be two to five t imes greater than those levels found outside. Even after these products have been used and then stored, high levels of the pollutants can remain in the air for long periods of time ( Indoor Air Quality 2010a ). These VOCs can cause Sick Building Syndrome ( SBS ) Symptoms of SBS are throat irritation, nausea, dizziness, and fatigue, to name a few (Indoor Air Quality 2010b). The EPA recommends a few things to reduce exposure and effects of VOCs and other forms of indoor air pollution. One way to reduce V OCs is to use the product in an open area with fresh air. U sing smaller quantities of these products will reduce the amount of indoor pollution. Limiting exposure is also a way to decrease the amount of indoor air pollution (Indoor Air Quality 2010a). C ontact with items that contain carcinogens, such as spray paint which releases methylene and fuels which releases benzene, shou ld be limited (Indoor Air Quality 2010a) Another way to decrease indoor air pollution is proper ventilation and using H igh Ef ficiency Particulate Air (HEPA) filters and air purifiers (ionic and ozone) in side the home (Indoor Air Quality 2010a and Absolute Air Cleaners, Air Purifiers, & Allergy Products 2010). Ionic air purifiers and ozone air purifiers are also sold. Ionic ai r purifiers em it negatively

PAGE 17

17 charged ions that attract dust, pollen, and other particles. These irritants are then pulled back in to the purifier by the purifie rs positively charged plates (Woolston 2008). However, these machines may not work as well as consumers have been lead to believe. Consumer Reports tested the Sharper Image Ionic Breeze and found that this machine does a poor job of removing smoke, dust, and pollen particles from the air (Consumer Reports 2007) Also, the ionic air purifier releases ozone which is harmful to people (Woolsten 2008). Ozone air purifiers are different from their ionic counterparts in that, instead of releasing negatively charged ions, these purifiers emit ozone molecules. Atoms are able to break off from the ozone molecule to affix themselves to other molecules An atom that has attached itself to another molecule transforms the chemical structure of the molecule ( Indoor Air Quality 2010c ). While manufacturers claim that this alteration can reduce the amount of indoor air pollution, ozone is harmful to people and can cause problems such as throat irritation and can exacerbate respiratory diseases Ozone also damages the lungs, causing reductions in lung function ( Indoor Air Quality 2010c ). In reviewing scien tific studies the EPA has found that ozone air purifiers do not have much effect in cleaning indoor air and can be detrimental to human health ( Indoor Air Quality 2010c ). In addition, t hese purifiers can take a long time to work, possibly months or even years, and do not remove some VOCs, such as carbon monoxide or formaldehyde ( Indoor Air Quality 2010c ). A safe and effective alternative to ozone and ionic air purifiers that does wo rk is indoor plants In 1989 Dr. B. C. Wolverton of the National Aeronautics and Space Administration (NASA), along with Anne Johnson and Keith Bounds, published a report entitled Interior

PAGE 18

18 Landscape Plants for Indoor Air Pollution Abatement. In this st udy it was found that indoor plants can reduce the levels of formaldehyde, benzene, and trichloroethylene in the air Indoor Plants take in these compounds through their leave s and breaks them down in the plants roots (Kobayashi et al. 2007) When these compounds are destroyed microorganisms in the soil use the remnants as food. The roots can absorb the compounds as well (Kobayashi et al. 2007). Indoor Plants that can be used to reduce indoor air pollution are, though not all at the same level of effectiveness: areca palm, bamboo palm, Boston fern, corn plant, dendrobium orchid, janet craig, warneckei, dragon tree, dumbcane, dwarf date palm, English ivy, Alii, florists mum, gerbera daisy, golden pothos, kimberley queen fern, king of hearts, lady palm, lily turf, peace lily, red emerald, rubber plant, schefflera, spider plant, and weepi ng fig (Kobayashi et al. 2007). While there is scientific research that these indoor plants can reduce indoor air pollution none of the studies have recommended market ing these indoor plants as air cleaners. Most of this research in the floricultural industry has focused on location and type of store, sales staff training, and advertising indoor plants for their aesthetic attributes. Research has also focused on tre nds and consumer preferences such as sustainability. Objectives There are two objectives with this study. The first objective is to determine consumers preference for indoor plants after being made aware that some indoor plants are able to remove indoor air pollution. The second objective is to determine how experimental designs in choice experiments can affect consumers valuation of goods.

PAGE 19

19 First Objective: Consumer Preferences One important question arises when consumers are given information: does the distribution of information (in this case the information about VOCs) affect consumers preferences and/or buying behavior? In this study some participants will be given information about VOCs and some participants will not. Receiving this information may change how a consumer views indoor plants and how they make purchasing decisions. Second Objective: Theoretical Implications of This Study This study used conjoint analysis which is a commonly used marketing tool. The conjoint us ed is choicebased conjoint (CBC) which allows participants to choose hypothetical products composed of different possible attributes and simulates a marketplace environment. This CBC is in the form of a survey emailed to the participants. Usually the at tributes are chosen by the researcher and are fixed for the survey. In this study some participants were given a set of attributes but other participants were allowed to choose the attributes that they prefer in indoor plants. By allowing participants to select the attributes they prefer more informative data can be collected and goods and services can be designed to better suit the needs of consumers. This research may show that allowing attribute selection may yield different, and possibly more informa tive, results than if participants were provided with just a fixed set of attributes. This approach could improve the way marketing data is collected.. Also, other research on floriculture has focused on areas such as plant containers and place of purchase. No other study has determined the attributes that consumers prefer in indoor plants. Sellers and growers can obtain valuable information on consumers opinions of indoor plants and the characteristics they would prefer these plants to have from this study.

PAGE 20

20 Figure 1 1. Total floriculture sales by producer from 2000 to 2010. Figure 1 2. Foliage sales by type from 2000 to 2010.

PAGE 21

21 Figure 1 3. Wholesale value of sales from 2000 to 2010. Figure 1 4. Number of Producers from 2000 to 2010.

PAGE 22

22 Figure 1 5. Existing home and condomi nium sales from 2006 to 2010. Figure 1 6. Median home prices from 2006 to 2010.

PAGE 23

23 CHAPTER 2 LITERATURE REVIEW The literature review will focus on two particular areas. The first section will discuss the ability of specific indoor plants to help clean indoor air. The second section will discuss the research on consumer preferences and marketing strategies Th e results of these studies help support the motivation of this study. Currently, there are no other studies available that have examined marketing indoor plants green. Studies on VOC Removal T he study by Wolverton, Johnson, and Bounds (1989) identifie d two issues concerning indoor air pollution: 1) determining if there are chemicals causing the pollution ; and 2) the relationship between these chemicals and SBS. The chemicals are released from equipment and furniture in e nergy efficient buildings This release leads to high levels of volatile organics in the air causing symptoms of SBS. The authors note another study b y Dr. Tony Pickering that investigated at the relationship between microorganisms and SBS. In buildings with natural ventilation and great amounts of these microorganis ms there are fewer occur rences of SBS symptoms than in buildings with mechanical ventilation and low levels of microorganisms. Pickerings findings led to the conclusion that there is no connection between microorganisms and SBS and that the cause of SBS is most likely volatile organics. Based on Pickerings results Wolverton, Johnsons and Bounds (1989) shifted the focus of their research to indoor plants. As people rely on plants for life there ar e beneficial attributes that can be exploited. Certain indoor plants can absorb pollutants commonly found in buildings. The three pollutants Wolverton, Bounds, and Johnson utilized we re benzene, trichloroethylene, and

PAGE 24

24 formaldehyde. Benzene is found in many household products such as gasoline, rubbe r, and in some paints. This pollutant can cause eye and skin irritation along with headaches and nausea In addition to being a carcinogen, benzene can cause respiratory diseases and kidney damage. Trichl oroethylene is found in items such as paints, varnishes, and inks. This compound can cause liver cancer. Formaldehyde is found in most buildings. It is also found in products such as household cleaners, tissues, grocery bags, and permanent press clothes This chemical can irritate skin and the membranes in the nose, eyes, and throat. It can also cause asthma and is suspected to be a major cause of a type of throat cancer Twelve different types of potted plants were used in this study : Bambo o palm, Chinese evergreen, English ivy, Ficus, Gerbera daisy, Janet Craig, Marginata, Mass ca ne, Mother in laws tongue, Peace lily, Pot mum, and Warneckei. Each plant was put in a sealed Plexiglas s chamber and then one of the three chemicals was released into the chamber. Once these chemicals were released t he roots of the plants absorb ed and broke down the chemicals and other organic pollutants. These broken down pollutants were then used to make new plant material. S oil for each plant was sealed to determi ne if it was helping to eliminate the chemicals. In addition to the plants and soil an activated carbon filter was used. The plants, in their original containers, were placed inside these filters. The reduction of pollutants increases when a carbon filte r is used as this filter removes smoke and organic chemicals. W hen reviewing the results Wolverton, Johnson, and Bounds found that the plants removed a great amount of the chemicals despite the low amount of lighting. Plants with f ull foliage were not as effective in removing benzene as those plants with more soil exposed to the air. Once the lower leaves were cut the plants removed more benzene. After adjusting for the

PAGE 25

25 full foliage the results showed that the plants and activated carbon filter s were effective in removing high levels of the three chemicals tested. The filters effectively trap ped the pol lutants and keep them until each plan ts roots and microorganisms were ready to use them as nutrients. This process is useful for reducing and r eusing carbon. Orwell et al. ( 2004 ) examined the rates of the removal of benzene by seven different types of pott ed plants. Kentia Palm, Peace Lily, Janet Craig, Sensation, Devils Ivy, Queensland Umbrella Tree, and another plant that is closely related to the Janet Craig. The plants were then put individual ly into sealed glass chambers. Benzene was then introduced into each chamber. After repeated samplings it was found that all the plants had similar and high, rates of benzene removal. Wood et al. ( 2006) investigated plants ability to eliminate VOCs based on prior research by Giese et al. ( 1994 ) Coward et al. ( 1996), and Lohr and PesrsonMims ( 1996) While the earlier research was conducted in labs Wood et al. (2006) utilized a nonlab environment. The authors used three buildings at the University of Technology Sy dney (UTS) in Australia. Two of these buildings wer e air conditioned and one used only natural ventilation. The first of two investigation s involved using a planting of a Janet Craig in each of the three buildings. A total of eighteen offices nine in the air conditioned building and nine in the naturally ventilated building, were used. The nine offices in each building were then divided into three subsets. In the air conditioned building t hree offices contained three plants; three offices contained six plants; and three offices contained zero plants. This process was repeated in the building with natural ventilation T he second investigation used the Janet Craig and the

PAGE 26

26 Peace Lil y Other than the addition of the Peace Lily this investigation was very similar to the first investigation T he authors identified fourteen VOCs in the offices. T he dominant VOCs were toluene, ethylbenzene and xylenes. These compounds can cause naus ea and loss of concentration in the short run. In the long run they can cause respiratory problems. The plants significantly reduced the amount of these VOCs in the offices in which they were placed. It was found that three Janet Craig plants were n eeded to clean the air ; increasing t he number of plants did not have any significant effect The authors also found there was no significant difference between how well the plants eliminated VOCs in the air conditioned buildings versus the naturally ve ntilated building. It was also determined that the plants did not interact with carbon monoxide concentrations that were present in the buildings. Oyabu et al. ( 2002 ) studied the effects of using potted plants to reduce VOCs in nursing homes in Japan using Golden Pothos and three types of potting soils. The VOCs tested were formaldehyde, acetone, and ammonia. The authors reported results of the purification capability for odor and VOCs. The purification capability (Pa) for odor took less time for som e chemicals than others. For example, the Pa acted slowly for ammonia because this chemical is used as nutrition for plants while the Pa for formaldehyde acted much faster because this VOC is absorbed by plant roots. Liu et al. ( 2006) reported the abili ty of seventy three species of ornamental plants to eliminate benzene. The authors conducted two trials, the first of which resulted in identifying plants that had eliminated more than 20% of the benzene. These plants were tested again in the second tria l to determine if they could absorb more benzene. The authors found that twenty

PAGE 27

27 three of the species did not change the amount of benzene in the air. Thirteen of the plants removed 19.99% of the benzene, seventeen removed 1020%, seventeen removed 2140%, and three removed 6080% of the benzene. It was also determined that some plants removed benzene quickly, some slowly removed the chemical, and some reduced benzene at a steady pace. Papinchak et al. ( 2009 ) examined how effective the snake plant, sp ider plant, and golden pothos are at decreasing ozone concentrations. The authors selected these plants as the y are common household plants and have been shown to help eliminate VOCs in previous studies. The authors found the time of day did not make a difference in terms of the rate of ozone removal and that t hese indoor plants can help to significantly reduce the amount of ozone in indoor environments. Wolverton and Wolverton ( 1993) tested over thirty plants to find how effective they are at rem oving formaldehyde and xylene, using both sterilized and unsterilized. The Boston fern was found to be the most effective at removing formaldehyde and xylene. T wo Boston ferns were more effective than three Janet Craigs in removing the chemicals The authors also found that plant leaf structure can help to remove two thirds to almost one half of both formaldehyde The results of the pots without plants were reported next The pots that contained only potting soil did remove a significant amount of formaldehyde but not as much as the pots containing the potting soil and Ficus b enjamina, Spathiphyllum sp., Sansevieria sp., and Kalanchoe sp. Matsumoto and Yamaguchi (2007) investigated how different types of light can affect foliage plants abil ities to remove VOCs. Four plants were used in the study : Benjamin, pathiphyllum Areca Palm and Concinna. Several different types of lights, light wavelengths,

PAGE 28

28 and brightness (illuminance) were used. The types of lights included fluorescent light, inc andescent light, and light emitting diode (LED). The brightness (illuminance) types included 350 1x, 700 1x, and 1050 1x. Light wavelengths included Red LED, blue LED and both red and blue LED. When examining Concinna and Spathiphyllum it was found t hat these plants remove toluene more efficiently under low luminance. Benjamin and Areca palm are more efficient when they are placed under lights with high luminance. All of the plants performed the best under the blue LED. These results may be helpful to consumers who are buying indoor plants and are concerned with maintenance. It is intere sting to note that while these studies demonstrate how specific indoor plants can remove indoor air pollution there is no research on how to market these indoor plants for this attribute. A few studies have investigated consumer preferences for attributes such as plant containers while others have examined the marketing strategies firms used. Consumer Preferences, Trends, and Marketing Strategies in the Floriculture Industry F ew studies have examined consumer preferences for different characteristics su ch as sustainability. Hall et al. (2010) investigated consumer preferences for biodegradable floral plant containers. The authors used conjoint analysis and surveyed 535 participants from four states. The study investigated w heat starch, rice hulls, and straw containers as alternatives to plastic containers and attributes such as the carbon footprint of these containers were specified in the survey. The authors also separated participants into seven distinct market segments: Straw

PAGE 29

29 Likers, Environ mentally Conscious, Price Conscious, Rice Hull Likers, Non discriminatin g, and Carbon Sensitive (Hall et al. 2010). Mason et al. (2008) examined consumer preferences for three attributes of container gardens: price, color harmony, and the quantity of care information given at the time of purchase. This study employed a webbased conjoint survey that included pictures of combinations of various plants in containers. From a sample of 985 participants the authors found that price was t he most important attribute, followed by quantity of care information and color harmony (Mason et al. 2008). Yue and Behe (2008) analyzed consumers choice of place of purchase for flowers. Five types of market outlets were considered: traditional freestanding floral outlets (TR); general retail stores (GR); box stores and mass merchandisers (BS); direct to consumers (DC) which includes internet based outlets; and other stores (O S). Two summary statistic s that are worth noting are that the majority of buyers were 40 years of age or older (73%) and that nearly 81% of buyers were women (Yue and Behe 2008). Yue and Behe used a multinomial logit whic h showed that when consumers bought flowers for themselves they tend ed to shop at BS but when these consumers purchased flowers for others as gifts, they tended to shop at TF. Also, when consumers made planned purchases they were more likely to buy at TF and DC but when they made unplanned purchases (e.g. impulse purchases) they were more likely to buy at GR, BS, and OS (Yue and Behe 2008). Consumers under the age of 25 and between the ages of 25 and 39 were more likely to buy flowers at DC than consum ers who were 55 years or older while c onsumers who were between the ages of 40 and 54 were more likely to buy from BS, DC, and OS than the other age groups

PAGE 30

30 (Yue and Behe 2008). When asked why they chose particular market outlets, participants who chose BS and GR did so because of the low prices and convenience offered. Partic ipants who chose TF cited the reputation of the TF, the quality of the products, and service. Participants who selected DC did so because of the convenience of delivery (Yue and Behe 2008). In addition to promoting indoor plants as air cleaners consider ation should be given to other methods of promotion in developing marketing plans. A new strategy may include promoting indoor plants as green and can be incorporated into current strategies. One area of a marketing plan is where floriculture products are sold. Wholesalers use marketing practices like trade shows and industry publications to reach large businesses (Hodges, Palma, and Hall 2010). However, these marketing strategies cannot be used when trying to target individual consumers. Consumers in this industry include homeowners and businesses such as resorts. Most floriculture products sold to these consumers are sold in through a variety of businesses such as florists, supermarkets, discount stores, mass merchandisers, and even flea markets (Hodges, Palma, and Hall 2010). The increase in retailers such as home improvement stores means large growers have access to a bigger market than before. These growers sell large amounts of floriculture products to these retailers or box stores (Hodges, Palma, and Hall 2010). Smaller or independent retailers are more likely to buy their products from independent growers. These retailers tend to have more experienced and/or trained sales staff and their customers are more likely to be interested in the q uality of the products rather than price (Hodges, Palma, and Hall 2010). While large box stores have the advantage sell ing their products at lower prices than do the independent stores but they do not often have trained sales staff. But box stores are be ginning to train their employees so that consumers can receive more information on the products they

PAGE 31

31 purchase (Hodges, Palma, and Hall 2010). Training the sales staff at the large box stores could greatly benefit both retailers and growers and s hould be considered as part of the marketing strategy. Independent stores often cannot lower their prices to compete with box stores. Instead some independent stores offer brands that are either their own or national. They provide classes or workshops on garde ning practices and some may also have cafes in their stores. These marketing practices help attract customers ((Hodges, Palma, and Hall 2010). Providing information to their customers about VOCs and the ability of indoor plants to remove them may help independent stores increase sales. Hodges, Palma, and Hall conducted a survey of 38,000 U.S. nurseries and asked these nurseries a series of questions about their marketing strategies. The findings showed that nurseries used the following marketing approaches: repeat sales, contract sales, and export sales with approximately 80% of nursery sales being repeat sales (Hodges, Palma, and Hall 2010). The nurseries also answered questions about advertising. It was found that approximately 4.6% of nursery sal es were spent on advertising through websites, radio and/or TV commercials, billboards, gardening periodicals, yellow pages, catalogs (either in print or online), trade periodicals and expos, newsletters, and other forms of advertising. Catalogs were the most utilized advertising tool followed by trade expos, other forms of advertising, and websites, respectively (Hodges, Palma, and Hall 2010). Through advertising nurseries can reach broader audiences and can inform consumers about the benefits of indoor plants. In addition, nurseries were also asked about factors that may influence how prices are determined. The factors included production costs, competitors prices, plant quality, plant

PAGE 32

32 uniqueness, quantity of plants in inventory, inflation, market dem and, and the previous years prices. The factors that most influenced how prices were set were production costs and market demand (Hodges, Palma, and Hall 2010). Moreover, market demand was identified as the most important issue that affected a nurserys business (Hodges, Palma, and Hall 2010). If nurseries are able to distribute information through advertising then market demand, the most important issue that influences their business, may increase. These studies show that although there has been res earch in the area of consumer preferences and marketing trends for floriculture there is more research that needs to be done. This study will expand on the work already completed and offer more insight into what consumers look for when purchasing plants. This study will also show how information can affect buying behavior and how possible marketing strategies can be employed to influence consumers purchases.

PAGE 33

33 CHAPTER 3 THEORETICAL FRAMEWORK In this chapter conjoint analysis is discussed. Data analysis will be discussed in Chapter 4 and Chapter 5. The current chapter uses six sources, Wittink and Bergestuen (2001), Vriens (1995), Green and Srinivasan (1978), Green and Srinivasan (1990), Wittink, Krishinamurthi, and Reibstein (1989), and from the anthology Conjoint Measurement: Methods a nd Applications, Second Edition (2001), as these sources have a wealth of information on how conjoint analysis is constructed. Guidelines and Principles of Conjoint Analysis According to Wittink and Bergestuen (2001) conjoint analysis is used to quantify how individual s confront trade offs when they choose between multidimensional alternatives. Researchers ask members of a target market to indicate their preferences (or choices) for objects under a range of hypothetical situations described in terms of product or service features, inc luding features not available i n existing products or services (147). From these choices researchers can determine a preference function for each individual. These preference functions can then be used to find part worths1 1 Vriens (1995) refers to part worths as part worth utilites or derived utility values or estimated regression coefficients or partial utility values for different attribute levels ( Wittink and Bergestuen 2001). From these part worths researchers are able to determine if consumers prefer one product t o other possible products. Thus, possible markets can be determined. Researchers can then predict market shares for companies and/or managers who would like to find new product/service opportunities ( Wittink and Bergestuen 2001).

PAGE 34

34 To better illustrate what conjoint analysis is Wittink and Bergestuen also provide a n example. The authors use a Proctor and Gamble (P and G) product, a diaper. The price is $5 for a dozen but has no elastic waistband. P and G is considering adding the elastic waistband and the possible prices for the altered product are $6 or $7. Here the attributes would be the el astic waistband and price. The levels for the elastic waistband are Yes and No. The levels for price are $5, $6, and $7. The researcher would then ask participants to rank each type of diaper that uses one combination of levels and attribute s, creating six possible combinations of the attributes ( Wittink and Bergestuen 2001). In a method known as the Adaptive Conjoint Analysis (ACA) (discussed more in the Data Collection Method: Adaptive Conjoint Analysis section ) the researcher presents th e combinations in pairs and participants chooses which combination of the two is preferred, ( Wittink and Bergestuen 2001). Instead of the ACA the researcher can provide the participants with a scale to rate their prefe rences. P articipants can rate each com bination on a scale of one to ten ( Wittink and Bergestuen 2001). From these preference scores the part worths will be constructed (this construction will be explained more in the next section). The authors list the main components of a conjoint study : the choice of a product category; the selection of a possible or target market; a determination of which attributes to include; the ranges of these attributes; a description of what the preference models may be and how the data will be collected; a descr iption of how the survey will be written; a description of how the survey will be conducted and how many consumers will receive the survey; and the analysis of the collected data ( Wittink and Bergestuen 2001).

PAGE 35

35 Certain conditions are needed in order for c onjoint analysis to be successful. First, participants in the study should be a representative sample. Second, participants must make the decisions for the product or service category. Third, the study should simulate an actual market environment so that participants will perceive the choices as realistic. In this situation participants will make decisions as they do in the marketplace. Fourth, different choices available should be represented using a small number of attributes. Finally, participants need to feel that there is a definite purpose to the study and therefore give honest responses ( Wittink and Bergestuen 2001). There are six principles that Wittink and Bergestuen give that should be used when conducting a conjoint analysis. The first pr inciple is simple and straightforward: conjoint analysis can accurately predict consumers purchasing behavior ( Wittink and Bergestuen 2001). The second principle states that the more complex a model is the better it will be at accurately forecasting mar ketplace behavior. At the individual level the validity of the model can be tested using the proportion of hits measure. This measure determines which hits (choices) are accurately predicted. However, companies and managers are more concerned with profi t shares and therefore are more concerned with validity at the aggregate level. The measure of validity at the aggregate level is the comparison of the share of predicted choices with the shares of choices of an alternative (or the holdout choice). Speci fically, researchers study the deviations between the two shares ( Wittink and Bergestuen 2001). A respondents preference function is used to predict holdout choices and its possible that this function may be misspecified. Bias may become an issue which can reduce validity. However, the true model has an error term which is asymptotically zero. The more complex the aggregate model is the more likely it is to be close the true model and bias will be reduced ( Wittink and Bergestuen 2001 ).

PAGE 36

36 The third pr inciple is the opposite of the second principle: a simple model can give better predictions than can a complex model at the individual level. At the individual lev el the more complex the model is, the greater the amount of unreliability. Bias is reduced but so is reliability. To control for this issue the researcher can limit parameter estimation ( Wittink and Bergestuen 2001). The fourth principle states that merging results from multiple methods can give better predictions than using just one method. When multiple methods are utilized each one provides an insight that is different from the others. These insights can improve forecasts ( Wittink and Bergestuen 2001). T he fifth principle is straightforward: if participants are given motivation then f orecasts will be more accurate. The authors cite previous research which has looked at ways respondents are motivated ( Wittink and Bergestuen 2001. The sixth principle may be more difficult to follow: the holdout task must be developed correctly so that the method used will reduce or eliminate bias and the forecasts will be more accurate than when other methods are used. Again the authors cite previous research as evidence for this principle ( Wittink and Bergestuen 2001). Theoretical Basis of Conjoint Analysis Vriens (1995) discusses two approaches to conjoint analysis, axiomatic deterministic conjoint measurement (ADCM) and numerical deterministic conjoint measurement (NDCM). ADCM was first developed in 1964 by Luce and Tukey. This method is used to characterize empirical systems of relations by a numerical system of relation s (Vriens 1995: 20). The author illustrates how this approach is utilized with a simple example. There are two attributes, p1 and p2, with levels l1 and l2 where l1, l2 = 1 2, or 3. J is the set of all possible combinations, in this case a total of nine combinations (the number of levels raised to the number of the attributes

PAGE 37

37 or 32 = 9). In ADCM the researcher cannot determine the utility of l1, l2 directly. The research er can find which combination is preferred over the others but not how much it is preferred. Instead participants can show preferences by ranking the combinations The ranked combinations can be expressed as y(l1 ,l2) in J (Vriens 1995). Three axioms ar e necessary for this approach: ordering, (single factor) independ ence, and double cancellation. The ordering axiom is weakly ordering and includes the following attributes: reflexivity, transitivity, antisymmetry, and conne ctedness. The independence axi om states that for all combinations in J: if y(1,1) < y(2,1) then y(1,2) < y(2,2) and if y(1,1) < y(1,2) then y(2,1) < y(2,2). T he double cancellation axiom states if y(1,2) < y(2,3) and y(2,1) < y(3,2) then y(1,1) < y(3,3) (Vriens 1995). After model testing number s are assigned to the levels. The numbers are assigned through scaling, a process where the inequalities in the axioms menti oned above are solved (Vriens 1995). The uniqueness problem, which entails finding a way to find how much freedom the researcher has in developing the scales can occur The set of admissible transformations of the measurement scales for which the model under study remains valid determines the measurement level of the measurement scales (Vriens 1995: 24). If the a xioms are met for a particular model then the scales will remain unchanged through linear transformations and will be considered to be interval scales (Vriens 1995 ). The other approach, NDCM, uses fit measures to determine if a particular empirical re lational structure used can be transferred to a numerical relational structure (Vriens 1995). Instead of testing the axioms, as what is done using the ADCM, the NCDM can be utilized.

PAGE 38

38 When using this method the model used is assumed to be the correct mode l and its fit is used as a me asure of its validity (Vriens 1995). The ADCM is also known as the deterministic approach because axioms are tested to find the correct model. Based on the results of these tests the model will be accepted or rejected. One common problem of the deterministic approach is that the data is prone to having errors. This problem indicates that any violation of the axioms may be caused not only by the wrong model but by data that contains errors (Vriens 1995). For the NCDM there are no distributional assumptions so the research can only look at the fit of the model to determine if it should be us ed (Vriens 1995). Selecting Attributes The first component of th e conjoint analysis is attribute selection The researcher must deter mine which attributes should be used, the number of attributes used, and how the attributes should be defined (Vriens 1995). The attributes and number included are often provided by companies and/or managers and consumers. Defining attributes, though, ca n be a difficult process. Physical attributes can be included as engineers can easily construct products from them. However, consumers may not be able to clearly interpret this type of at tribute so the researcher must describe the attributes in terms of benefits to consumers (Vriens 1995). The second step is defining the levels of the attributes. Specifically, the researcher must determine the variation of the levels and the number of levels. Vriens cites an unpublished paper by Vriens and Wittink (1992) to help with these issues. The authors found that if an attribute is continuous then the range of that attribute should extend from the minimum amount to the

PAGE 39

39 maximum amount that is possible for a particular produc t. Including more intermediate lev els can help the researcher collect more informative data. One disadvantage with increasing the number of intermediate levels is the attribute level effect ; the relative importance of the attrib utes is affected by the number of intermediate levels (Vriens 1995). If the researcher is comparing two attributes the relative importance of one attribute with three levels may be higher than the other attribute with two levels. The researcher may conclude that the difference between the highest and lowest levels for the former attribute is more significant than the difference of the highest and lowest levels of the latter attribute (Vriens 1995). Vriens describes three solutions if the researcher encounters the attribute level effect. The first solution is to have the same number of levels for each attribute. However, holding attribute levels constant may not be possible in practice as some attributes may require two levels and other attributes may have more than two (Wittink, Krishnamurthi, and Reibstein 1989). The second solution is to use a higher measurement scale which can induce participants to provide a least interval scale measurement. The third solution is to use more self explicated data where the number of levels should correlate with self e xplica ted importances (Vriens 1995). Wittink, Krishnamurthi, and Reibstein (1989) discuss four more options for the attribute level effect. The authors named this problem the comparability problem. The solutions for the comparability problems are based on the importance of each attribute. These importances are determined using maximum likelihood estimation to demonstrate the difficulty of comparison when the attribute levels are not constant. The first option is to modify the attribute importances based on systematic differences in the possible values between attributes varyi ng in the number of levels (Wittink, Krishnamurthi, and Reibstein 1989). However, there is no clearly defined

PAGE 40

40 best way to adjust these results. The second solution is to aggregate the importances and then compare them. One approach is to determine how often each attribute is most important (Wittink, Krishnamurthi, and Reibstein 1989). The third option is to disregard the concept of importances and compare attribute effects on market shares if the comparability problem only occurs for the importances an d not the shares (Wittink, Krishnamurthi, and Reibstein 1989). The fourth option offered is to use rating scales to determine preferences (explained in more detail in the Data Collection section ). With this option the problem of comparability does not ex ist (Wittink, Krishnamurthi, and Reibstein 1989). The third component to be considered is how the attributes are presented. Vriens suggests four ways to represent attributes: 1) verbal; 2) pictorial; 3) both verbal and pictorial; and 4) actual produc ts. There are advantages and disadvantages to both verbal and pictorial descriptions. The advantage of a verbal description is that the construction is relatively easy but creating a pictorial representation may be a more difficult process (Vriens 1995). In some situations, though, the pictorial description may be more advantageous than the verbal. If the attribute is a design attribute then a picture would be a better way to show the attractiveness of a product. The verbal representation may also be m ore abstract than a picture. A more realistic description such as a picture can help simulate a real market environment and participants may provide more honest and realistic answers (Vriens 1995). Data Collection Gustafsson, Herrmann, and Huber ( 2001 ) st ate that the first step of data collection is to c hoose the preference function (not selecting attributes). The three types of preference functions are the (ideal) vector model, the ideal point model, and the partial benefit model (also called the

PAGE 41

41 part wo rth s function model). However, Vriens considers model selection as the first step in data analysis, discussed in the Data Analysis section In this study the data collection method will be the first stage of data collection. Vriens (1995) discusses four methods: 1) the full profile method; 2) the tradeoff matrix approach; 3) the method of paired comparison; and 4) the adaptive conjoint analysis procedure. Additionally, choice based conjoint analysis can be used. Data Collection Method: t he Full Profile Method The full profile method requires participants to evaluate a group of hypothetical products that differ from each other by a least two different attributes (Vriens 1995). All attributes and attribute levels are used in this design, also called the complete or factorial design (Gustafsson, Herrmann, and Huber 2001). There are advantages and disadvantages of this method. One advantage is that the participants must see the stimuli that contain all attributes and make decisions based on these attributes. This situation can lead to a marketplace environment where participants may behave as they would in an actual market. The second advantage is that this method may bring about interaction effects. The third advantage is the possible increase in flexibility (Vriens 1995). One disadvantage is that the participants can be given too much information and too many tasks. The excess information is a result of the number attributes and the complexity of the stimuli. Consequently, the number of attributes and the amount of stimuli creates an overload of tasks and participants may feel overwhelmed (Vriens 35). The full profile method may be time consuming for the participants; therefore, another design can be used. The reduced design, or a fractional factorial conjoint, is composed of a subset of the combination of attributes and levels (Gustafsson, Herrmann, and Huber 2001). There are two ways to achieve this design: 1)

PAGE 42

42 random sampling (not used in marketing research); and 2) reduce the complete design to so that orthogonality is maintained (Gustafsson, Herrmann, and Huber 2001). Data Collection Method: t he Tradeoff Matrix Approach The tradeoff matrix approach ( called the two factor method by Gustafsson, Herrmann, and Huber (2001) and by Green and Srinivasan (1978)) is used to reduce the number of tasks associated with the fullprofile method. Participants are given tradeoff matrices and then asked to rate or rank each combination of attributes in the matrix (Vriens 1995). One disadvantage is that the number of matrices increases as the number of attributes increases. If there are p attributes there will be (p(p 1))/2 matrices (Vriens 1995). For example, if there are two attributes there will be 1 matrix whereas if there are 10 attributes there will be 45 matrices. Another disadvantage is that in an actual marketplace products do not consist of only two attributes. This lack of realism could potentially inval idate results (Vriens 1995). Data Collection Method: t he Method of Paired Comparisons The method of paired comparisons presents two hypothetical products to the participant who then indicates the product he or she prefers. In this scenario the products can include some or all of the attributes (Vriens 1995). The advantage of the method of paired comparisons is that participants are more likely act the way they would in a real marketplace. The disadvantage is that if the number of hypothetical products increase so do the number of paired comparisons. As with the tradeoff matrix approach if there are p profiles then the number of paired comparisons will be p(p 1)/2 (Vriens 1995). For example, if there are 10 profiles there are 45 paired comparisons for p articipants to evaluate.

PAGE 43

43 D ata Collection Method: Adaptive Conjoint Analysis The adaptive conjoint analysis (ACA) is similar to the method of paired comparisons. However, to keep the analysis less complicated the participant is asked to give self expli cated assessments of the attributes (Vriens 1995). This data is then used to determine an initial solution and based on this solution the first paired comparisons are selected by a computer. Two products having to two to five attributes each are shown to the parti cipants. The participants then rank how much they prefer one product over the other using a nine point scale. Once the participant has ranked the two products the initial solution is revised and the second pair of products is chosen in a simila r manner as the first pair (Vriens 1995). Each pair is chosen so that the predicted level of preference is nearly equal. This method is effective as it does not ask participants to make decisions in which the outcome would be obvious (Vriens 1995). Data C ollection Method: ChoiceBased Conjoint Analysis Choice based conjoint (CBC) analysis allows the participant to choose a specific product among many products. This situation imitates a marketplace environment (Sawtooth Software 2008). As with the other conjoint methods there are advantages and disadvantages. One advantage is that this method also permits the participant to choose none of the possible products. Another advantage is that interaction effects can be estimated. An additional advantage is that product attribute levels can include either product or alternative specific levels (Sawtooth Software 2008). This method also has disadvantages. Often too many product choices are included in the analysis which can be overwhelming to the parti cipant s Also, CBC analysis results in less data than with other conjoint methods. This problem occurs because the

PAGE 44

44 products are not ranked against each other. One solution to these issues is to reduce the number of attributes (Sawtooth Software 2008). Data Collection Method: Two Other Approaches Green and Srinivasan (1990) suggest two more approaches for handling large numbers of attributes. The first approach is the self explicated approach which requires participants to rate levels of attribute o n a scale. This desirability scale may be from 0 (least desired) to 10 (most desired) with all other attributes held constant. The participant assigns points among the attributes so that the relative importance of each attribute is represented. This relative importance is called an importance weight. The researcher will then estimate the part worths by multiplying these weights by the ratings chosen with the desirability scale (Green and Srinivasan 1990). While the advantage of this approach is that it is rel atively easy to implement there are several disadvantages such as intercorrelation between attributes. Participants may not be able to accurately rate attribute levels as i t may be difficult for participants to hold all other attribute levels constant. Another disadvantage is that there may be biases when collecting socioeconomic characteristics. A third disadvantage is that an additive part worth model is assumed to be the true model. The data from full profile rankings is fit to an additive model. When the model is multiplicative then it can be made additive by a logarithmic transformation if estimatio n is nonmetric. Misspecification is not as much as a problem with the fullprofile approach as it is with the self explicated approach (Green and Srinivasan 1990). A fourth disadvantage is that if there are redundant attributes the information collected may also be redundant. The participant may not realize this redundancy in the self explicated approach but

PAGE 45

45 he or she may be able to avoid it in the full profile approach. A fifth disadvantage is that the self explicate approach may display only linearit y in the part worths. The full profile method has the ability to display both linearity and nonlinearity in the part worths The sixth and final disadvantage is that the researcher cannot collect data on the probability the participant will purchase a particular product (Green and Srinivasan 1990). The second method Green and Srinivasan (1990) recommend is the hybrid method. This approach uses both self explicated data and a full profile conjoint. The self explicated data is collected to determine an i nitial set of partworths. The full profile conjoint contains a small number of profiles taken from a large set of profiles for participants to assess. These subsets are extracted so that all the profiles are assessed by different subsamples of participants at market segment levels. When multiple regression is used market segment level adjustments to partworths (and, if desired, interaction effects) are estimated by relatin g ... the overall preferences f or the full profiles to the self explicated utilities. Each respondents self explicated utilities then can be augmented by segment level parameters estimated from the multiple regress ion (Green and Srinivasan 1990: 11). The advantage of the hybrid approach is that issues with the self explicated app roach are reduced. Also, the amount of information from the full profile approach is decreased, thereby easing the participants workload (Green and Srinivasan 1990). Fractional Factorial Conjoint and Choice Based Conjoint Analysis: A Comparison Based on the information provided thus far in this chapter, survey construction can be discussed further. As mentioned in the Data Collection Method: Choice Based Conjoint Analysis subsection, in CBC analysis a marketplace environment is simulated. This environment is created so that participants feel like they are in an actual market making realistic purchasing

PAGE 46

46 decisions (Sawtooth Software 2008). In contrast, the fractional factorial conjoint uses a subset of the profiles from the full profile method. Because this type of conjoint uses fewer attribute level combinations or runs there is a possibility that the effects can be confounded. In order to avoid this problem the fractional factorial conjoint can be designed to be orthogonal (SAS Institute Inc. 1993). This design, an orthogonal array, ensures that the effects are uncorrelated. The orthogonal design can be used when there are a small number of both attributes and levels (as there are in this study) (SAS Institute Inc. 1993). To determine which design should be used its efficiency measures can be measured and compared. There are two types of efficient measures. The first is A efficiency. This efficiency is based upon the arithmetic mean of the eigenvalues. The A efficiency takes the forms of A efficiency = (3 1) where trace ((XX)) 1/p is the arithmetic mean and p is the number of attribute levels (SAS Institute Inc. 1993). The other type of efficiency measure is D efficiency. This measure is based upon the geometric mean of the eigenvalues. This efficiency is D efficiency = (3 2) where | XX1 |1/p is the geometric mean (SAS Institute Inc. 1993). Fractional factorial conjoints can be designed with main effects o nly or with main effects and twoway interaction effects (Lusk and Norwood 2005). If the researcher chooses the design with only main effects the quantity of profiles needed decreases. The main effects are orthogonal to each other. If the researcher sel ects the design with both main effects and twoway effects the

PAGE 47

47 effects will be orthogonal but the number of profiles will be greater than with the main effects only design (Lusk and Norwood 2005). Lusk and Norwood note that even though the fractional fact orial design can reduce the number of profiles there may still be too many profiles for participants to examine. Therefore a CBC can be used. Evaluation of the Attributes To evaluate the attributes the participants are usually provided a rating scale wh ich is a metric scale level (Gustafsson, Herrmann, and Huber 2001). With this scale participants rate the potential benefits of each hypothetical product on a numbered scale. This type of evaluation permits the researcher to collect data that is ordinal (Gustafsson, Herrmann, and Huber 2001). A rating scale is sometimes preferred to a ranking scale as it shows the strength of the participants preference. The ranking scale is ordinal as well but there is one disadvantage. The ranking scale is nonmetric and can only show that one hypothetical product is preferred to another, not how much it is preferred. (Gustafsson, Herrmann, and Huber 2001). The other nonmetric scale is paired profiles comparison. As discussed in the Data Collection section, in this situation the researcher provides the participant with two hypothetical products at one time. Then the participant makes a decision as to which product he or she prefers. This scale is advantageous in that it can help the participant avoid intransitiv ities that can occur with graded paired comparisons (Gustafsson, Herrmann, and Huber 2001). Selection of the Data Collection Procedure Three common ways that conjoint analysis data is conducted include inperson interviews, computers, and a combination of phone interviews and mail (Gustafsson, Herrmann,

PAGE 48

48 and Huber 2001). The in person and computer data collection are two methods that help to ensure that the data collected will be informative and accurate (Vriens 1995). Conducting mail surveys alone may be low cost but not as effective as other methods. Its possible that the researcher may not be able to tell who answered the survey. Also, the participant may not be able to contact the researcher in case he or she needs help with the survey (Vriens 1995). A way to address the problems with the mailonly survey is to conduct a combination of mail and telephone surveys. Participants are chosen randomly or through another type of method commonly used in telephone surveys and then survey materials are sent to these participants by mail. After a specified amount of time the researcher calls the participant to obtain preference measures (Vriens 1995). In this study, however, the author will use the internet and email the surveys to participants. Data Ana lysis According to Vriens (1995) there are four steps in the data analysis: 1) model specification; 2) choosing the estimation method; 3) evaluation and selection of the model; and 4) market simulations. The data collection stage can influence how each step in the data analysis is done (Vriens 50). Data Analysis: Model Specification The model specification step involves selection of the attributes, which was explained in section and expands on the discussion by Gustafsson, Herrmann, and Huber (2001) i n the Selecting Attributes section. Vriens includes a discussion on what kind of constraints should be

PAGE 49

49 used with these attributes. In the following subsections the constraints will be discussed as will the preference functions associated with each. With in attributes constraints The first type of constraint is the within attributes constraint. These are constraints on the part worths of the various levels of each attribute (Vriens 1995). There are three model options for withinattribute constraints. The first is the part worth function model also called the partial benefits model by Gustafsson, Herrmann, and Huber ( 2001) which determines the utility an individual receives from a particular product that may have many attributes. The model is: yij = p p=1 uip (ljpij (3 3) where Uij = p p=1 uip (ljp) (3 4) and i, j, p, ljp,, and yij, are as they were described in the Theoretical Basis of Conjoint Analysis section. Uij is utility and ij is the error term. This model does not have the within attributes constraint on the main effects and has the greatest level of flexibility of all the models (Vriens 1995). In this model for every level of an attribute a value, or part worth, is assigned to it. Green and Srinivasan (1978) descr ibe the model slightly differently: sj = t p=1 fp(yip) (3 5) where yip is a level of an attribute p in the jth stimulus set ( vip in Vriens notation) and fp is a function of this level. One condition of this model is that a linking rule is used to determine the

PAGE 50

50 total benefit of an alternative good from the partia l benefit values found by using the preference function (Gustafsson, Herrmann, and Huber 2001). The second model option for model specification is utilizing linear constraints on the estimates of uip (ljp). This model is called the vector model. The vector model permits either ordinal or interval constraints. If ordinal constraints are imposed the n the part worth function model is used with the following inequalities uip (1jp) < uip (2jp) < uip (3jp) ... < uip (Ljp) (Vriens 53). If the interval constraints are imposed the model is slightly different. The vector model takes the form: yij = p p=1 wip vjp ij (3 6) where wij is the weight an individual places on a particular attribute and vjp is the value of the levels of a particular attribute (Vriens 1995). There is a linear relationship between the levels and their resulting part wor ths. The independent variable has to be evaluated on an ordinal scale (Vriens 1995: 54). The third model option is the idealpoint model. In this model quadratic constraints are imposed on uip (ljp). This model is similar to the vector model. The model is: yij = p p=1 wip ( vjp Oip)2 ij (3 7) where Oip is the ideal value of a particular attribute, ( vjp Oip)2 is the squared distance between the two values, and wip is the weight an individual places on that distance. This weight must be less than zero (Vriens 1995: 54). As ( vjp Oip)2 increases the levels part worth utility

PAGE 51

51 decreases. In this model the attributes must be measured on an interval scale. The idealpoint model also shows that there is a quadratic relationship between the levels and their resulting partworths (Vriens 1995). Flexibility of the partworth function model is useful for the withinattribute constraints. Green and Srinivasan (1978) demonst rate how the part worth function model can be converted to both the vector and idealpoint models. When fp(yjp) is set equal to wp(yjp xp)2 (where wp is the weight and xp is the ideal value, Oip in Vriens notation) the part worth function model is tran sformed into the idealpoint model. And when fp(yjp) is set equal to wpyjp the part worth function model is transformed into the vector model (Green and Srinivasan 1978). The ideal vector model will appear if all the manifestations of the attributes are c onsidered to be dummy variables and the number of parameters to be estimated is minimized. The amount of parameters to be estimated for the ideal point model varies between the number of the parameters estimated for both the ideal vector model and the par tial benefit model (Gustafsson, Herrmann, and Huber 2001). Across attributes constraints The second type of constraint discussed is across attributes constraints. These constraints are used to deal with attribute main effects and interaction effects. When this type of constraint is not used then a fully saturated model is to be estimated. A fully saturated model is one that contains all main and interaction effects and can only be used if the full profile design is utilized. For reasons mentioned in subsection Data Collection Method: The Full Profile Method this design is not easy to implement. The difficulty of implementing the fullprofile design ensures that a constraint is placed on the parameter estimates (Vriens 1995). The two types of across -

PAGE 52

52 attribute constraints are zero one across attribute constraint and the ordinal inequality across attribute constraint. Each is discussed in the subsections below. Zero one across attribute constraints The zero one across attribute constraint allows the researcher to set a few main effects to zero or a few of the interaction effects to zero (Vriens 1995). In some studies all interaction effects are set to zero, leaving only the main effects. If there are some higher orde r interaction effects they are set zero. However, including some interaction effects is necessary for two reasons. One reason is that these interaction effects can display some consumer behavior. Disregarding these effects can mislead companies and mana gers about what type of product should be produced. The second reason is that including interaction effects can increase the power of prediction of the conjoint (Vriens 1995). If the interaction effects are necessary the researcher must determine what the first order interaction effects are. The researcher can ask the manager, who might know which attributes interact. The researcher can use a pilot study to determine the possibility of interaction effects. Also, the researcher may have some theoreti cal basis on which he or she believes interaction effects occur (Vriens 1995). When examining interaction effects Vriens discusses hierarchal models as nonhierarchal models are seldom used. Hierarchal models take the form of: yij = p p=1 uip (ljp) + p p=1 p q=p+1 ui,pq (lj,pqij (3 8) where lj,pq is the l th level of the interaction of the attributes, p and q, ui,pq (lj,pq) is the utility of lj,pq, and yij uip, ljp, and ij represent the same values as before (Vriens 1995). Two types of first -

PAGE 53

53 order interactions effects should be considered. The first type is the crossover interaction effects. This type of effect occurs when the preference ordering for the levels of one attribute (e.g. p ) are dependent on the level s of another attribute (e.g. q) (Vriens 1995). The second type of interaction effect is the noncrossover interaction effect. This type of effect occurs when the utility functions of each attribute are affected by the levels of another but the relationsh ip is such that the preference for the levels of other attributes is not changed (Vriens 1995). The ability to estimate interaction effects depends on how the data is collected, the characteristics of the design of the conjoint, and how the dependent vari able is measured (Vriens 1995). For some of the data collection methods interaction effects cannot be measured. The tradeoff matrix approach does not allow for interaction effects because these effects are confounded with (Vriens 1995). The ACA does a llow for interaction effects but this type of conjoint analysis has been modified to permit only main effects. The ACA and the tradeoff matrix approach can only include interaction effects for attributes that can be aggregated to one large attribute. For the method of paired comparisons and the full profile approach interaction effects can be measured (Vriens 1995). The conjoint design can affect how interaction effects can be estimated. If main effects only models are used interaction effects can confou nd them. If a small number of interaction effects are included this problem may not occur. If it is not possible to exclude interaction effects compromise designs can be used. Compromise designs include all the main effects and some interaction effects (Vriens 1995). These main effects and interaction effects must be mutually uncorrelated (Vriens 1995). Compromise designs can cause problems. If there are attributes that have three or more levels and there are interaction effects between the attributes

PAGE 54

54 the participant will given at least thirty two profiles. Blocking designs can be used instead of compromise designs. In a blocking design groups of participants are provided with different types of profiles. These profiles are taken from a pool of prof iles, or a master design. As with the compromise designs there is a disadvantage with blocking designs. Individual level models cannot be estimated (Vriens 1995). The measurement of the dependent variable can affect how and if interaction effects are estimated. If the data is ordinal level data some types of estimation, such as ordinary least squares (OLS), can produce noncrossover interaction effects. These effects can disappear if a transformation of the data takes place. The multivariate analysis of variance (MANOVA) is one method to transform the data monotonically. This nonmetric procedure is the only procedure that can perform this transformation and remove the noncrossover interaction effects (Vriens 1995). If both crossover and noncrossove r interaction effects are required to be present then the dependent variable should be evaluated on a metric scale (Vriens 1995). Ordinal inequality across attribute c onstraints The second type of across attribute constraint is the ordinal inequality acro ss attribute constraint. This constraint is placed upon the relative importances of each attribute. However, this constraint has one caveat. If the researcher has prior information about the importances then this constraint can used but if there is no prior information this constraint cannot be used. The predictive ability of the conjoint increases when this constraint is placed on the attributes and the attribute level effect decreases (Vriens 1995).

PAGE 55

55 Across subject constraints The final type of constraint is the across subject constraint. This constraint is imposed on the part worths across subjects. If the model is estimated at the individual level (as researchers sometimes design their experiments to allow for this estimation) then there are no across subject constraints (Vriens 1995). If the researcher decides to use the model for the entire sample then across subject constraints can be used to set the part worths equal to each other. The parameters are set equal for all subjects because th e preference ratings are grouped across participants. These parameters are the same as the average of the parameter estimates that would be obtained for all individuals. One problem that might arise is the majority fallacy. This problem occurs when the heterogeneity that a sample has is lost when the parameters are estimated (Vriens 1995). Another way to apply this constraint is use it among subgroups. Within the subgroups the parameters must be the same but they are permitted vary among the groups. F our partitioning schemes that can be employed are non overlapping, overlapping, fuzzy, and factor analytic (Vriens 1995). The nonoverlapping scheme requires that the sample is divided into subgroups that do not overlap and parameters are estimated for ea ch subgroup. The overlapping scheme requires that the sample be put into subgroups but participants can be included in more than one subgroup. Parameters are then estimated for each subgroup. The fuzzy scheme allows the researcher to place participants into many subgroups but only partially. Again parameters are estimated for each subgroup. The factor analytic scheme permits the participants to be described by factors not clusters and these factors determine which constraints are used. The parameters are then estimated for the factors (Vriens 1995).

PAGE 56

56 Estimation Method The next step in data analysis is the selection of the estimation method. The researcher must consider the scaling of the dependent variable. The evaluation of the dependent variable c an also be done on a measurement scale. As with evaluating attributes this scale can be metric or nonmetric though the variable is nonmetric itself. Although the dependent variable is measured on a nonmetric scale the estimated betas have properties o f the interval scale (Green and Srinivasan 1978). The evaluation of the dependent variable can either be in terms of overall preference or how likely it is that a consumer will buy the product (Green and Srinivasan 1978). If the scaling is non metric the dependent variable is thought to be ordinal scaled. Conversely, if the scaling is metric then the dependent variable is thought to be least intervalscaled (Vriens 1995). Advantages exist for each type of scaling. More information may be available when the data is collected using metric scaling. Non metric scaling has three advantages. The first advantage is that ranked data will be more consistent. This consistency occurs because the respondent is stating a preference between two goods. The second advantage is that the part worth functions can be added or multiplied. The third advantage is the tradeoff matrix approach ranking of cells in the table doesnt rely upon the levels of possible missing factors. Metric scaling, however, is affected by missing factors (Green and Srinivasan 1978). In terms of results neither type of scaling is advantageous over the other. If the researcher collects nonmetric data OLS should not be used. This does not necessarily mean that nonmetric estimation has better or more predictive power than metric estimation (Vriens 1995).

PAGE 57

57 The researcher must also consider that by imposing constraints the estimation will be affected. In other words estimation methods can be characterized by the possibilities to impose 1) within attribute constraints, 2) across attribute constraints and 3) acrosssubject constraints (Vriens 1995: 65). For example, MANOVA may be used for a nonmetric dependent variable and if the researcher wants to employ across attribute constraints he or she can only consider crossover interactions effects (Vriens 1995). Evaluation and Selection of the Model The third step in data analysis is to evaluate and select the model. The measure of fit is one approach to determine how well the model is suit ed to the data and should be done with caution for three reasons (Vriens 1995). The first reason is data from conjoint designs frequently provide high goodness of fit outcomes. This result occurs often if models are estimated for individual participants. The second reason is that many models can give measures of fit that are similar. The third reason is how to determine how the outcome of a test of significance should be used and what value should be considered significant (Vriens 1995). Another appr oach to determine if the model is the correct model is to examine the measures of reliability and the predictive power of the model. Vriens discusses an example where participants are given a set of eight possible products and select the product they pref er. After this task they are given a second task with the same eight possible products where they again choose the product they prefer (sometimes called a holdout sample). The researcher can examine the first choice each participant has made and predict which products the participants will select when presented with the same choices. The researcher can also obtain estimates from the first task and predict the holdout sample. If the model estimates accurately predict the

PAGE 58

58 choices the participants actually make in the holdout sample the researcher can determine how well the model predicts participants choices (Vriens 1995). The holdout sample can include products that are not used in the estimation sample. Also, the predictive power of the model increases when the design of the holdout task is designed to simulate a market environment (Vriens 1995). Market Simulations The fourth and final step of data analysis is market simulations. Often the manager wants to forecast market shares. To do this forecast ing the researcher must convert the part worth estimates into choice estimates, which must be transformed into aggregated preference estimates. Several models are available so the researcher must evaluate each one to determine which will best fit the data (Vriens 1995). Three commonly used models that transform part worth estimates into choice estimates are the first choice model, the Bradley Terry Luce (BTL) model, and the logit model. The f irst choice model is either : Cij prob = { 1 iff Uij = maxh Uij; h = 1,...,I (3 9) or { 0 iff Uij < maxh Uij; h = 1,..., I (3 10) The (BTL) model is: Cij = Uij / J j=1 Uij (3 11) The logit model (used in this study) is:

PAGE 59

59 Cij prob = eUij / J j=1 eUij (3 12) Vriens designates a matrix U which contains Uij (72). Cij prob is the function assigning choice probabilities to the elements of U (Vriens 1995: 72). The first choice model may not be the appropriate model for the data for two reasons. The first reason is that the researcher wi ll expect that the product with the highest predicted utility will have the highest actual utility. This assumption may cause a problem as in some situations the participants responses increases as the number of products increases and there is little var iation in their utilities. The error term, ij is zero (Vriens 1995). Issues occur with the BTL and logit models as well. The part worths are sometimes measured on an interval scale level. Since they are measured on this scale they can be altered using linear transformations. These transformations, if BTL and logit models are used, can have an effect on market share predictions (Vriens 1995). Including a constant can influence the BTL choice probabilities. Multiplying the partworths can change the l ogit choice probabilities. Also, if there are substitutes for the choice set the researcher may not be able to use the logit and the BTL models as they may be inadequate (Vriens 1995). A simulated probit may be used to deal with these issues (Vriens 1995). Tests of Reliability and Validity Tests of validity and reliability were mentioned briefly in the Data Analysis section and are explained more thoroughly here. T he tests of reliability can be done with either parameters or participants. O ne test of reliability discussed by Green and Srinivasan (1978) is a test of participants input judgments and is similar to the test discussed by Vriens (1995). The

PAGE 60

60 researcher can ask a subsample of participants to evaluate a subset of the original set of stimuli or hypothetical products. The researcher can then determine if the participants choices are the same. In between the original set and the subset the participants should be given some sort of task to complete (Green and Sr inivasan 1978). Another test of reliability is done with parameters. Again a subsample of participants evaluates a second set of stimuli or hypothetical products. However, this second set is not a subset of the original stimuli. The researcher could t hen estimate product moment correlations from the parameters from both tasks. This estimation is called the coefficient of equivalence (Green and Srinivasan 1978). The first test of reliability using just the participants only takes into account possible errors with input data. The second test of reliability examines errors with input data, the variability of how the hypothetical products are set up, errors in how the parameters are estimated, and possible variations in time periods (Green and Srinivasan 1978). The two types of validity that the researcher can test for are internal validity and external validity. Internal validity can be demonstrated by using tests of correlation such as Pearsons or Spearmans rho. This rho measures the correlation between inputted dependent variable and estimated dependent variable (Green and Srinivasan 1978). The data can also be used to test validity across two sets. If the researcher conducted a test of reliability with a subsample of participants and a subset of stimuli then a determination can be made of whether the first set of data accurately predicts the second set (Green and Srinivasan 1978). The external validity test used to measure how accurately actual participants responses match predicted response s is the predicted validity test. The researcher can use the estimated preference function to determine the participants predicted rankordered choices. The

PAGE 61

61 participants actual choices should be on the rank order from 1 to n, where n is the number of s timuli or hypothetical products. The researcher can construct a frequency distribution of the rank order choices. This distribution is compared to the uniform distribution to test for statistical significance and the test statistic used is the Kolmogorov Smirnov (Green and Srinivasan 1978). In Chapter 4 the construction of the conjoint and the data collection procedure used in this study will be explained. The selection of attributes used and their levels will be discussed. Statistics and demographic data from the surveys will also be shown.

PAGE 62

62 CHAPTER 4 METHODOLOGY In Chapter 3, the two methods used in this study, the fractional factorial conjoint and the choicebased conjoint, were described. In this chapter the construction and the distribution of the surveys will be discussed. In addition, preliminary results of the analysis will be shown. Construction of the Surveys The next issues to consider are how many attributes, levels, and choice sets should be included. Two attributes that had to be included were Price and VOC Removal. Five levels of Price ranging fr om $15 to $55 in increments of $10 were used. These levels were chosen as they were similar to the prices of plants that remove VOCs. The second attribute required was the two levels of VOC Removal (remove and does not remove). For this attribute there c ould only be these two levels as indoor plants either remove VOCs or do not remove VOCs. To determine which attributes should be included in addition to Price and VOC Removal focus groups were conducted in Gainesville and Jacksonville, Florida in October of 2010. The participants of the focus groups were asked a series of questions such as how often they buy plants, where they buy these plants, what attributes they prefer in plants, and how different attributes affect their purchasing decisions. When as ked which attributes were most important in a houseplant participants indicated that Sunlight Needed, Hardiness, and Flowering were the three most important (aside from Price and VOC Removal). The majority of the participants were concerned that their hom es did not let in enough light for indoor plants to survive. Participants were also concerned that indoor plants required a lot of care and they did not have the time to maintain a plants needed level of care. Many of the participants were interested in a

PAGE 63

63 houseplant that was hardy and did not require much care. Also, a number of the participants preferred indoor plants that are attractive and flowers increase a houseplants attractiveness for many participants. Height, Toxicity, and Tags identifying the plants were considered to be important, though not as important as Sunlight Needed, Hardiness, and Flowering. Height was important as some participants did not want a houseplant that would take up a lot of space in their homes. Many of the participa nts have children and/or pets and were concerned about indoor plants being accidentally ingested. Tags were mentioned by some participants as important because they are often not sure of the name of the plant or what the plant will look like at maturity. All of these attributes had either two or three levels (Table 4 1). Sunlight Needed levels included full/direct sunlight, partial sunlight, and little/indirect sunlight. Hardiness levels included needs a lot of care, needs some care, and needs little care. Height had three levels: grows to 2 to 4 feet at maturity; grows to 4 to 8 feet at maturity; and grows to 8 to 12 feet at maturity. Flowering, Toxicity, and Tags had two levels: one with the attribute and one without. From these attributes and their levels hypothetical indoor plants were determined. For example if a participant chose the attributes Height, Hardiness, and Flowering one hypothetical houseplant may be $15, grows to 2 to 4 feet at maturity, needs little care, is flowering, and does remove VOCs. Height, Toxicity, and Tags were not selected by participants but all six of the attributes were used in the choice based conjoint (CBC) for this study, though not every attribute were given to the participants.

PAGE 64

64 In traditional conjoint a nalysis a fixed set of attributes are provided for the participants. This study is different from traditional conjoint analysis in two ways. First, for the CBC constructed for this study, approximately 1/3 of the participants were given a fixed set attributes, Sunlight Needed, Hardiness, and Flowering, as is most frequently done in conjoint analysis. However, slightly more than 2/3 of participants were allowed to choose three of the six attributes that they preferred in a houseplant. This selection was done to determine which attributes are most important to consumers. For example a participant may select Height, Hardiness, and Flowering as the attributes he or she may want a houseplant to have. By permitting attribute selection it can be determined if the attributes chosen by the focus groups are the attributes that are most important to other consumers. Additionally, more can be learned about the possible different market segments that have different attribute preferences. If this is not the case th en allowing participants to choose the attributes they prefer may be a better approach to CBC. The total number of possible combinations of these attributes that participants were able to choose from was 20. Second, if participants were given three of the six attributes discussed by the focus groups and the attributes VOC Removal and Price in the full profile method the amount of information would have been overwhelming, as discussed in Chapter 3. To prevent participants from feeling inundated with thi s information, a subset of the full profile was used. This fractional factorial design was reduced again when using when using a CBC as this method limits the number of profiles that a participant receives even further. Once the data collection method was determined the choice sets and choices were constructed. Each of the 20 combinations of attributes contained choice sets and each choice set

PAGE 65

65 had several hypothetical indoor plants or choices. From each choice set the participants chose the housepla nt he or she most preferred. Within each choice set the choices are unique and are not repeated in the other choice sets. When testing the number of choice sets in SAS it was found that for each of 20 combinations of attributes, 12 choice sets with 5 pos sible choices or hypothetical indoor plants in each set gave the highest D efficiency. Each time 12 was used as the number of sets the D efficiency was at least 97 or above. Employing more or fewer sets reduced the D efficiency. The choice sets were con structed using SAS marketing macros. The macros utilized were %mktruns, %mktex, %mktlab, and %choiceff. These macros ensured that the design was orthogonal as well as efficient. Once the choice sets were determined using SAS the surveys were then des igned in Qualtrics, a survey software program. For the CBC that permitted attribute selection each of the 20 combinations of attributes was one branch of the survey that contained 12 choice sets with 5 choices in each set. For example if the participant chose Flowering, Tags, and Sunlight Needed as the attributes he or she preferred in a houseplant, the participant would be directed to a branch with the 12 choice sets, each containing 5 hypothetical plants or choices composed of the levels of the attribu tes Flowering, Tags, Sunlight Needed, VOC Removal, and Price. The participant would then choose one of the hypothetical indoor plants from each of the 12 choice sets. If the participant instead chose Tags, Height, and Toxicity as the attributes that were most important to him or her, the participant would be directed to the branch with the 12 choice sets of hypothetical indoor plants composed the of the levels of the attributes Tags, Height, Toxicity, VOC Removal, and Price. For the participants that rec eived the fixed set of attributes there was only one branch, which contained 12 choice sets created from attribute levels of Sunlight Needed, Hardiness, Flowering, VOC Removal, and Price.

PAGE 66

66 Additionally, to test the impact of information about VOCs on par ticipants choices of indoor plants participant were assigned, at random, to a survey with or without information about how certain indoor plants reduce them indoors. This randomization permitted comparisons between participants choices based on how much they knew about VOCs. It is expected that the inclusion of VOC information will have an impact on the choices of those participants who receive it. Preliminary Results Participants were recruited using Survey Sampling, Inc. Validation questions were used to ensure that participants were reading the survey carefully. Also, participants were required to have an annual household income of at least $35,000 to participate in the survey. A total of 2280 surveys were completed. Descriptive statistics for both surveys allowing attribute selection will be discussed first. There were 1,555 respondents to the survey with attribute selection and 725 for the surveys with fixed attributes. Table 4 2 displays how often participants purchase indoor plants. The participants who received a survey with attribute selection purchase indoor plants with nearly the same frequency as those participants who were given the fixed set of att ributes. The majority of participants buy indoor plants once a year and 24% never purchase indoor plants. For those who purchase indoor plants the most common place to buy them are home improvement stores (Table 43). The second most popular type of store to purchase indoor plants is discount stores. For those participants who purchase indoor plants the most common reason is for decoration. The reason given by the majority of those participants who do not purchase indoor plants is that they

PAGE 67

67 have a hard time keeping these plants alive. The majority of participants (60%) do not have trouble finding the variety of houseplant they want (Table 44). Men accounted for 48% of participants for the surveys with attribute selection versus 49% for the surveys with fixed attributes (Table 4 5). Table 46 displays the ethnicity of the participants. Most of the participants are Caucasian in each type of survey. The youngest participant was 18, the oldest was 85 or over, and the average was 46 (Table 47). As shown in Table 4 8 the majority of the participants did not have children less than 18 years of age living in the home. Many of the participants had pets: dogs and cats were the most common though participants from both surveys had a variety of pets (Ta ble 4 9). Most of participants from both the surveys with attribute selection (79%) and the surveys with fixed attributes (78%) were homeowners and most lived in single family homes (Table 4 10 and Table 4 11). The participants in both surveys tend to be primary shoppers for the household (Table 4 12). Sixty three percent of participants from both surveys were married. Single participants accounted for 22% and 23% of all participants from the surveys with attribute selection and the surveys with fixed attributes, respectively (Table 4 13). Participants tended not to have respiratory problems (Table 414) and the most common allergies were pets, medicines, indoor and outdoor allergies (Table 415). Table 4 16 displays the income distribution of the pa rticipants. The participants had to have an annual household income of at least $35,000 to qualify for the study. Few participants had annual income above $150,000.

PAGE 68

68 In Chapter 5 the analysis of the data used will be explained. Conditional logits were used to determine why respondents chose specific indoor plants. These logits analyze the choices based on attributes and their levels.

PAGE 69

69 Table 4 1. Attributes and their corresponding levels Attribute Level s Price $15, $25, $35, $45, $55 Sunlight Little/indirect, Partial, Full/direct Hardiness Needs little care, needs some care, needs a lot of care Height Grows to 2ft to 4ft, grows to 4ft to 8ft tall, grows to 8ft to 12 ft tall Tags Has a tag clearly identifying the plant, does not have a tag Toxicity Plant is toxic, plant is not toxic Flowering Plant is flowering, plant is not flowering VOC Removes VOCs, does not remove VOCs Table 4 4 Problems finding plant variety Have Trouble Attribute selection (%) Fixed attributes (%) Total (%) Yes 8% 7% 8% No 60% 61% 60% Sometimes 32% 32% 32% Table 4 2 Frequency of purchasing indoor plants (attribute selection) Frequency Attribute selection (%) Fixed attributes (%) Total (%) Twice a Month 4% 4% 4% Once a Month 7% 6% 7% Every Few Months 19% 20% 19% Every Six Months 13% 13% 13% Once a Year 23% 23% 23% Once Every Five Years 9% 9% 9% Never 24% 24% 24% Other 1% 1% 1% Table 4 3 Place of purchase (can purchase at multiple stores) Store type Attribute selection (%) Fixed attributes (%) Total (%) Home Improvement Stores 66% 62% 65% Discount stores 43% 44% 43% Grocery Stores 22% 26% 23% Nurseries 38% 37% 38% Large Garden Stores 18% 17% 18% Small Garden Stores 19% 20% 20% Other 3% 3% 3%

PAGE 70

70 Table 4 5 Gender of participants Gender Attribute selection (%) Fixed attributes (%) Total (%) Male 48% 49% 47% Female 52% 51% 53% Table 4 6 Ethnicity of participants (select all that apply) Ethnicity Attribute selection (%) Fixed attributes (%) Total (%) African American 5% 5% 5% Asian 4% 3% 4% White 89% 88% 89% Hispanic 2% 4% 3% Native American or Alaska Native 1% 1% 1% Pacific Islander 0% 0% 0% Other 1% 2% 1% Table 4 7 Ages of participants Age Attribute selection Fixed attributes Minimum 18 18 Mean 46 46 Maximum 85+ 85+ Table 4 8 Number of children living in household (select all that apply) Age of Children Attribute selection (%) Fixed attributes (%) Total (%) Under 2 years old 8% 7% 8% 2 5 years old 15% 12% 14% 6 11 years old 15% 15% 15% 12 17 years old 19% 19% 19% No children living in the household 61% 63% 62% Table 4 9. Pets in living in the household (select all that apply) Pet Attribute selection (%) Fixed attributes (%) Total (%) Dog 50% 49% 49% Cat 40% 36% 39% Bird 63% 6% 6% Hamster, gerbil, ferret or similar pet 4% 3% 3% Snake, lizard, gecko, iguana, or similar pet 3% 4% 3% Other 7% 6% 7% None 27% 3% 28%

PAGE 71

71 Table 4 10. Ownership of home Own or Rent Attribute selection (%) Fixed attributes (%) Total (%) Own 79% 78% 78% Rent 21% 22% 22% Table 4 11. Type of home Building Attribute selection (%) Fixed attributes (%) Total (%) Apartment 12% 12% 12% Single family home detached from any other structure 76% 76% 76% Duplex 2% 3% 3% Townhouse 6% 5% 6% Other 4% 4% 4% Table 4 12. Primary shopper in the household Primary shopper Attribute selection (%) Fixed attributes (%) Total (%) Yes 91% 90% 94% No 9% 10% 10% Ta ble 4 13 Marital status of participant Marital status Attribute selection (%) Fixed attributes (%) Total (%) Single 22% 23% 22% Married 63% 63% 63% Separated 2% 1% 2% Divorced 8% 7% 8% Widowed 3% 3% 3% Never Married 2% 2% 3% Table 4 14 Respiratory problems Have respiratory problem Attribute selection (%) Fixed attributes (%) Total (%) Yes 13% 11% 12% No 87% 89% 91%

PAGE 72

72 Table 4 15. Allergies (select all that apply) Type of Allergies Attribute selection (%) Fixed attributes (%) Total (%) Pets 10% 13% 11% Foods 7% 7% 7% Medicines 15% 14% 15% Indoor allergies, e.g. dust 20% 24% 21% Outdoor allergies, e.g. hayfever 35% 37% 36% Plants 4% 4% 4% Others 3% 3% 3% None 51% 49% 51% Table 4 16 Annual household income Income Attribute selection (%) Fixed attributes (%) Total (%) $35,000 $49,999 29% 28% 29% $50,000 $74,999 38% 38% 38% $75,000 $99,999 16% 17% 17% $100,000 $149,999 11% 13% 12% $150,000 $199,999 4% 3% 4% $200,000 $249,999 1% 1% 1% $250,000 + 1% 0% 1%

PAGE 73

73 CHAPTER 5 RESULTS In Chapter 4 the design of the survey, data collection, and preliminary results were discussed. In this chapter the analysis of the data will be discussed. This analysis will include the regressions used to determine why participants chose specific indoor plants. Attributes Selected by Participants Though Sunlight, Hardiness, and Flowering were the attributes chosen b y the focus groups, when attribute selection was permitted participants chose different combinations (Table 51). The combinations most frequently selected were Sunlight, Hardiness, and Height; Sunlight, Hardiness, and Flowering; Sunlight, Hardiness, and Toxicity; and Sunlight, Hardiness, and Tags. This finding is significant and supports the second objective. Attribute selection may play an important role in conjoint analysis as consumers have differing ideas about which attributes products should have. Companies and researchers may consider this method in addition to focus groups as other consumers may have different views and opinions from those who are interviewed in these groups. This approach can give researchers better insight to consumer prefere nces. Regression Analysis The conditional logit was used to analyze the participants specific plant choices. The conditional logit was first developed in 1974 by Daniel McFadden from random utility functions (Maddala 1983). The random utility function i s composed of two parts: 1) the nonrandom component which is the maximum amount of utility that a consumer can obtain given constraints such as income and the other possible alternatives, and 2) the random component that includes

PAGE 74

74 unobserved factors including consumer preferences and the characteristics of the alternatives (McFadden 1980). This logit, like the multinomial logit (MNL), analyzes the probability of one choice among other choices. The MNL analyzes choice probability using the attributes of the participants; however, the conditional logit analyzes this probability based on the characteristics of the choices (Maddala 1983). The MNL is usually shown as: J j j i m i i im y1) exp( ) exp( ) | Pr( x x x (5 1) where Pr(yi = m | xi) is the probability of a specific outcome m xi is the characteristics of the individual respondents, and j = 1,...,J are the possible outcomes (Long 1997) In general, the conditional logit takes the form of: J j ij im i im y1) exp( ) exp( ) | Pr( z z z (5 2) where Pr( yi = m| zi) is the probability of a specific outcome m zi is the characteristics of the product that respondent i prefers, and J is the total number of products in a particular choice set (Long 179). In this study, i is the respondent, j is the plant chosen, and zi is the plant attributes and levels that the respondents have selected. If the participant received the survey where he or she was able to select attributes, zi may include Sunlight Needed, Height, and Hardiness with their associated levels. Willingness to pay (WTP) for each attribute were calculated as well. WTP is as follows:

PAGE 75

75 P jWTP (5 3) where j is the coefficient of attribute j and P is the coefficient of the price attribute (Ryan and Hughes 1997). This WTP is how much more (or less) a participant would pay for a houseplant to have a specific attribute. These estimates can also be int erpreted as the amount a participant would pay for the attribute itself (Ryan and Hughes 1997). Table 565 shows the weighted means of WTP for all attributes except for Price. Conditional Logit Results As mentioned in the previous section the conditional logit was used to determine why participants selected specific indoor plants. This process was done for all surveys whether the participant received a survey with VOC information and attribute selection, a survey with VOC information and attribute selection, a survey with fixed attributes and VOC information, or a survey with fixed attributes and without VOC information. For clarification the variables Flowering, Tags, Toxicity, and VOC removal each have two levels, or are binary variables. Flowers (the variable name for Flowering), Tags, and Toxic (the variable name for Toxicity) was assigned a 1 if the plant flowered, had a tag, or was toxic and a 0 if not, respectively. VOC (the variable name fo r VOC removal) was given a 1 if the plant removed indoor air pollution and a 0 if the plant did not clean indoor air. VOC (the variable name for VOC removal) was given a 1 if the plant removed indoor air pollution and a 0 if the plant did not clean indoor air.

PAGE 76

76 The attributes Sunlight Needed, Hardiness, and Height each have three levels. The variable SunlightP is assigned a 1 if the plant needed partial sunlight and 0 if not. The variable SunlightFD is assigned a 1 if the plant need full/direct sunlight and 0 if not. SunlightLI is the reference category, indicating that the plant needs little/indirect sunlight. The variable Hardy1 is a 1 if the plant needs some care and a 0 if not. The variable Hardy2 is a 1 if the plant needs a lot of care and a 0 i f not. The variable Hardy0 is the reference category, or that the plant needs little care. Height4 is given a 1 if the plant will reach 4 to 8 feet at maturity and a 0 if not. Height8 is given a 1 if the plant reaches 8 to 12 feet at maturity. Height0 is the reference category and is 1 if the plant grows to a height of 2 to 4 feet at maturity and 0 if not. The variables SunlightP, SunlightFD, Hardy1, Hardy2, Price, Toxic, Height4, and Height8 are expected to have negative signs. SunlightP and Sunli ghtFD are each expected to have a negative sign because many households may not have enough natural light coming to the home. Hardy1 and Hardy2 indicate that the participant would have to spend more time maintaining the plant and many of the participants may want to spend their time do other activities. Price is assumed to be negative because it is believed that participants would rather spend less money on a houseplant than more. Toxic is assumed to be negative as some households have small children or pets and may worry that these children and pets might accidentally ingest a poisonous plant. Its possible that some of the respondents may be concerned that they themselves might accidentally ingest the plant. Height4 and Height8 are also expected to be negative because these plants take up a lot space. Some participants may not find this attribute attractive, especially if they have small homes and/or rooms.

PAGE 77

77 The variables Tags, Flowers, and VOC are expected to have positives signs. Tags is assumed to be positive as some participants may want to see indoor plants that have tags that give them detailed information about the houseplant. Flowers is expected to be positive as the focus group participants indicated they preferred a houseplant to be flowe ring. VOC is expected to be positive even though only about half of the respondents were given information about VOC. Some of the participants may have heard about VOCs and that certain indoor plants can remove them. Also, it is possible that respondent s may like seeing that indoor plants can do something. The results for the surveys allowing attribute selection will be discussed first. Next the results for the surveys with fixed attributes will be discussed. A goodness of fit measurement will als o be included in the discussion. One goodness of fit measurement is McFaddens likelihood ratio index. This measurement is similar to the R2 statistic in ordinary least squares. The likelihood ratio index is calculated as: 0 2ln ln 1 L L RM (5 4) (SAS Institute Inc. 2011). L is the log likelihood function at its maximum and L0 is the log likelihood function when the parameters are constrained to zero. As with the R2 in ordinary least squares the 2 MR is between 0 and 1 (SAS Institute Inc. 2010). After the results for the surveys with attribute selection are explained the results for the surveys with fixed attributes will be discussed. Finally, there will be a comparison of all the results.

PAGE 78

78 Surveys with Attribute Selection The results of the conditional logits and WTP estimates for the attributes selected as most important to participants are shown in Table 52. Significant variables had the expected signs except in four cases: Tags for the combination of Sunlight, Height, and Tags (without VOC information); Flowers for the combination of Hardiness, Toxicity, and Flowering (with VOC information), Hardy1 in Sunlight, Hardiness, and Toxicity (with and without VOC information); and VOC for the combination of Sunlight, Height, and Tags (without VOC information). With the exception of Sunlight, Hardiness, and Toxicity all of these combinations had low likelihood ratio indices. A summary of the significance and signs of the variables are shown in Table 53. VOC and Price were included in all combinations of attributes. Without information VOC was significant in all 20 of the regressions; with information this parameter was significant in 19. Price was significant in 18 of the logits when information was pr ovided; when information was not provided Price was significant in 19 of the logits. VOC, when significant, was positive except for Sunlight, Height, and T ags Price, when significant, was always negative. Significance will be discussed later in the cha pter. Surveys with Fixed Attributes The results for the conditional logits and WTP estimates for the plants with fixed attributes (Sunlight, Hardiness, and Flowering) are shown in Table 54. In both regressions (with and without information) the parameter s all have the expected signs and are significant except SunlightP. The coefficients for Price, SunlightFD, and Flowers are larger when there was

PAGE 79

79 no VOC information provided. The coefficients for Hardy1, Hardy2, and VOC are larger when there was VOC info rmation provided. Discussion and Comparison of Willingness to Pay for Attributes VOC removal All parameter estimates for VOC were significant except for Hardiness, Toxicity, and Flowering with VOC information. With one exception (Sunlight, Height, and Ta gs) WTP for VOC was positive for each group of participants when participants did not have information about VOCs. The WTP estimates for VOC increased when VOC information was given in all but three combinations: Sunlight, Tags, and Toxicity; Height, Har diness, and Tags; and Sunlight, Height, and Toxicity. This implies that providing the participants with information about VOCs and how specific indoor plants can remove them had a positive effect on purchasing behavior. Flowering and tags The coeffici ent for Flowers was included in 9 sets and was always positive, though for Hardiness, Toxicity, and Flowering with information it was negative. When combined with Tags and Toxicity with no information provided the coefficient was not significant. With the exception of the combination Hardiness, Toxicity, and Flowering the WTP estimates for Flowers decreased when VOC information was provided to participants. The attribute Tags was not significant for four combinations: Flowering, Tags, and Toxicity, with and without information; Flowering, Tags, and Hardiness, with information; Height, Tags, and Flowering, with and without information; and Height, Tags, and Toxicity with

PAGE 80

80 information. In all but two cases (Hardiness, Tags, and Toxicity, and Sunlight, Tag s, and Toxicity) the WTP estimates for Tags also decreased when VOC information was provided to participants. Participants who received VOC information were willing to pay more for the VOC removal, offsetting the amount they were willing to pay for Flowers or Tags. It is also possible that Toxicity may have had an effect on the WTP for Tags and Flowers as this att ribute was included in each of these attribute combinations where WTP did not increase when information was given. Toxicity Toxic was significant and negative for all combinations except for Hardiness, Flowering, and Toxicity when information was provided where it was not significant. The WTP for Toxic was always negative though this attribute experienced both increases and decreases in WTP when VOC information was given. The WTP for the attribute Toxic increased when the information was provided for t he following seven combinations of attributes: Flowering, Tags, and Toxicity; Height, Tags, and Toxicity; Height, Flowering, and Toxicity; Sunlight, Toxicity, and Flowering; Sunlight, Tags, and Toxicity; Sunlight, Hardiness, and Toxicity; and Sunlight, He ight, and Toxicity. Participants may have been more willing to purchase a plant that was toxic if it removed VOCs as this benefit may have outweighed concerns about toxicity. WTP for Toxic decreased for Hardiness, Tags, and Toxicity and Hardiness, Height and Toxicity when information was given.

PAGE 81

81 Sunlight SunlightP (the indoor plant requires partial sunlight; SunlightLD, the reference category, indicates that the plant requires little/indirect light) was only significant for two combinations when VOC i nformation was not given: Sunlight, Height, and Tags; and Sunlight, Hardiness, and Flowering. When VOC information was given SunlightP was significant for only Sunlight, Hardiness, and Tags. In each of these cases the coefficient SunlightP was negative SunlightFD was not significant for Sunlight, Height, and Tags when information was not given and was not significant for Sunlight, Height, and Toxicity when information was both provided and not provided. This coefficient was significant in all other cases. When significant SunlightFD was always negative. The WTP for SunlightFD (an indoor plant requiring full/direct sunlight) increased for five cases: Sunlight, Height and Flowering; Sunlight; Toxicity, and Flowering; Sunlight, Tags, and Toxicity; S unlight, Hardiness, and Toxicity; and Sunlight, Hardiness, and Flowering when VOC information was provided. This WTP decreased for four combinations of attributes: Sunlight, Tags, and Flowering; Sunlight, Height, and Tags; Sunlight, Hardiness, and Tags; and Sunlight, Hardiness, and Height. Hardiness The parameter estimates for Hardy1 (an indoor plant requires some care; Hardy0, the reference category, indicates the plant requires little care) though generally negative, were not significant for most of the regressions, implying that even if the participants were given the VOC information they were indifferent towards this attribute level. For three cases where these parameters were significant (Sunlight, Hardiness, and Flowering; Sunlight, Hardiness, and Tags;

PAGE 82

82 and Sunlight, Hardiness, and Height) the increases in WTP for Hardy1 when information was provided were positive though the WTP estimates remained negative. Sunlight is the common attribute in these three combinations and its presence may have an effect on Hardy1 when VOC information is given. Though Hardy1 was not significant in many of the logits, Hardy2 (the plant required a lot of care) was significant (and negative) in most of the regressions with the exception of Hardiness, Toxicity, and F lowering with no information provided. The WTP for Hardy2 increased when VOC information was given with the exception of Hardiness, Tags, and Toxicity, and Height, Hardiness, and Toxicity. Participants who were given the VOC information may not have want ed the indoor plant to need a lot of care but this may have become less important to them when they learned about some plants ability to remove VOCs. This information may have influenced how the participants viewed other attributes, especially VOC remova l. Height Height4 was not significant in the following combinations: Height, Tags, and Flowering, with information; Height, Toxicity, and Flowering with information; Height, Tags, and Toxicity without information; Hardiness, Height, and Toxicity without information; and Sunlight, Height, and Tags without information. When Height4 was significant the WTP estimates were mostly negative but increased when VOC informant was given. Height8 was significant and negative in nearly every case except for Sun light, Height, and Tags when information is not given. Participants were more concerned about the houseplant

PAGE 83

83 if it would grow to 8 to 12 feet at maturity whether they had the information or not. The WTP estimates for both Height4 and Height8 were almost always negative but most increased when VOC information was provided. The exception to this was for Height8, where the WTP had decreased when information was provided and the combination of attributes included: Height, Tags, and Flowering; Height, Hardiness, and Toxicity; and Sunlight, Height, and Tags. There is no attribute that is common among the three combinations of attributes that might offer an explanation for this decrease. Fixed attributes The results for the fixed attributes surveys showed t hat the WTP for VOC removal did increase and the WTP for Flowers decreased when VOC information was provided. The WTP for SunlightFD, while negative for both groups of participants, decreased slightly; most of the changes in WTP for SunlightFD where attri bute selection was allowed were much greater. The WTP for Hardy1 and Hardy2 were negative when VOC information was not provided and when it was provided; however, WTP for both Hardy1 and Hardy2 increased when information was provided. Attribute selection and fixed attributes To compare the use of attribute selection with the fixed attribute approach the results of the logits when Sunlight, Height, and Flowering were selected should be analyzed. For the participants who received the surveys allowing attri bute selection and chose these three attributes SunlightP was significant only in the regression without VOC information and its WTP estimate was $3.24. In the fixed attribute model SunlighP was not significant, with or without

PAGE 84

84 information. The WTP for SunlightFD, though negative, increased from $15.07 to $11.05 when information was provided for the participants who selected attributes. For the fixed attribute model SunlightFD was again significant both with and without information and increased from $8.85 to $8.21. For the attribute selection model WTP for Hardy1 increased from $9.18 to $4.34 and Hardy2 increased from $34.90 to $25.27 when information was provided. When attributes were fixed Hardy1 did increase but was negative when information was not provided ( $7.93) and when it was provided ( $5.66). Hardy2 was negative as well when information was not provided and when it was provided but increased from $34.95 to $24.92 when information was present. WTP for the attribute VOC removal di d increase when information was given but the overall WTP was smaller for the participants who were in the survey with attribute selection. The WTP for VOC for those participants who received the fixed attribute surveys was $23.59 and $41.04 without and w ith information, respectively. This WTP for the participants who selected these attributes was lower in both cases: $15.55 and $30.33, without and with information, respectively. Ranges and Weighted Averages of Willingness to Pay for Each Attribute The ranges of WTP for each attribute and/or level are shown in Table 55. The WTP ranges shown were calculated from conditional logit estimates that were significant. For example, Sunlight (Partial) was significant in only one regression without VOC informat ion and in only one regression with VOC information. The fixed attribute surveys did not include the attributes Height, Tags, and Toxicity and are not included in these WTP ranges. The weight ed

PAGE 85

85 averages of WTP for the attributes can provide more insight into how much participants were willing to pay for each attribute. W eighted means of the WTP for all attributes except for price were calculated (Table 56) by weighting each WTP estimating by the percentage of participants who chose each attribute. For example, if Hardiness was selected then all the WTP estimates for Hardy1 and Hardy2 were multiplied by the percentage of participants who chose this attribute. The weighted means were calculated for surveys allowing attribute selection, both with and without VOC information. WTP values were excluded from the weighted averages under two conditions: wh en they were chosen by less than 1% of the participants ; and estimates derived from logits with low likelihood ratio indices as the values may be not accurate measures of WTP. Some of the WTP estimates had high values and may have been much higher than the actual value of the plant. This result was not expected and may have occurred because of hypothetical bias (Lusk and Schroeder 2004) Participa nts were not purchasing an indoor plant so the WTP estimates, and the WTP weighted means could overstate act ual WTP. Though some of the WTP estimates may have been affected by hypothetical bias, the weighted means can be used to represent how much consumers are willing to pay for each attribute especially compared to the other attributes.

PAGE 86

86 Table 5 1 Number of participants per survey attribute selection Attributes selected No VOC information (number) No VOC information (%)* VOC information (number) VOC information (%)** Hardiness, tags & flowering 22 3% 13 2% Hardiness, toxicity & flowering 21 2.5% 18 2% Hardiness, toxicity & tags 11 1% 27 4% Height, hardiness & flowering 28 3% 22 3% Height, hardiness & tags 16 2% 11 1% Height, hardiness & toxicity 15 2% 23 3% Height, sunlight & flowering 94 11% 56 7% Height, sunlight & hardiness 109 13% 97 12.5% Height, sunlight & tags 64 8% 59 8% Height, tags & flowering 14 2% 8 1% Height, toxicity & flowering 10 1% 6 1% Height, toxicity & tags 8 1% 7 1% Sunlight hardiness & flowering 92 11% 88 11% Sunlight, hardiness & toxicity 95 11% 82 11% Sunlight, hardiness & tags 84 10% 89 12% Sunlight, height & toxicity 38 4.5% 44 6% Sunlight, tags & flowering 43 5% 35 4.5% Sunlight, toxicity & flowering 40 5% 34 4% Sunlight, toxicity & tags 22 3% 42 5% Toxicity tags & flowering 7 1% 7 1% *These percentages are for surveys with no VOC information only; **these percentages are for surveys with VOC information onl y

PAGE 87

87 Table 5 2. Parameter estimates and WTP for flexible attribute models Parameter No VOC information VOC information Estimate T value WTP Estimate T value WTP Flowering, tags, and t oxicity Price 0.060*** 5.64 n/a 0.070*** 6.11 n/a Flowers 0.195 0.62 $3.24 1.112*** 3.53 $15.93 Tag 0.494 1.56 $8.20 0.495 1.6 $7.09 Toxic 1.405*** 3.79 $23.34 1.416*** 4.14 $20.28 VOC 1.036*** 3.07 $17.21 1.990*** 5.51 $28.52 LR index 0.29 0.33 Flowering, tags, and h ardiness Price 0.049*** 8.74 n/a 0.066*** 8.17 n/a Flowers 1.019*** 5.88 $20.62 0.762*** 3.4 $11.60 Tag 0.338* 1.76 $6.83 0.117 0.49 $1.78 Hardy1 0.600*** 3.35 $12.15 0.192 0.79 $2.91 Hardy2 2.144*** 7.13 $43.40 0.996*** 3.27 $15.16 VOC 0.463*** 2.68 $9.37 1.081*** 4.58 $16.45 LR index 0.24 0.24 Hardiness, toxicity, and flowering Price 0.061*** 8.39 n/a 0.008 0.4 0 n/a Hardy1 0.335 1.44 $5.46 0.592*** 2.83 $347.94 Hardy2 1.301*** 4.3 $21.18 0.107 0.62 $62.82 Toxic 2.065*** 7.07 $33.63 0.142 0.92 $83.71 Flowers 0.923*** 3.99 $15.03 0.386** 2.42 $227.18 VOC 1.111*** 4.53 $18.01 0.221 1.43 $ 130 .00 LR index 0.42 0.03 Hardiness, tags, and toxicity Price 0.066*** 6.69 n/a 0.042*** 7.23 n/a Hardy1 0.401 1.52 $6.09 0.294* 1.84 $6.94 Hardy2 0.787** 2.33 $11.95 1.996*** 7.83 $47.18 Tag 0.598** 2.09 $ 9.07 0.556*** 3.12 $13.14 Toxic 1.450*** 5.03 $22.00 1.723*** 9.51 $40.73 VOC 1.372*** 4.37 $20.82 1.836*** 8.83 $43.41 LR index 0.29 0.29 H eight, t ags and f lowering Price 0.047*** 7.34 n/a 0.07 0*** 6.04 n/a Height4 0.469** 2 .00 $10.03 0.368 1.09 $ 5.25 Height8 0.658*** 2.75 $14.06 1.896*** 3.76 $27.08 Tags 0.203 1.09 $4.33 0.124 0.42 $1.78 Flowers 1.088*** 5.12 $23.24 0.780** 2.2 0 $11.15 VOC 0.677*** 3.59 $14.46 1.250*** 3.77 $17.85 LR index 0.18 0.33 *, **, and *** indicate significance at the 90 % 95 % and 99% confidence level s, respectively

PAGE 88

88 Table 5 2. continued Parameter No VOC information VOC information Estimate T value WTP Estimate T value WTP Height, t oxicity and f lowering Price 0.008 1.3 0 n/a 0.030*** 2.96 n/a Height4 0.680*** 2.71 $87.14 0.402 1.2 0 $13.64 Height8 1.158*** 4.02 $148.50 1.183*** 2.77 $40.08 Toxic 0.864*** 3.84 $110.82 1.545*** 4.31 $52.36 Flowers 0.972*** 4.07 $124.55 0.949*** 2.66 $32.16 VOC 0.404*** 1.89 $51.82 1.869*** 5.23 $63.36 LR index 0.14 0.28 Height, tags, and t oxicity Price 0.024*** 2.69 n/a 0.029 *** 2.71 n/a Height4 0.387 1.21 $16.32 0.822** 2.13 $28.05 Height8 0.940** 2.51 $39.65 1.905*** 3.93 $65.02 Tags 0.530* 1.83 $22.35 0.271 0.82 $9.26 Toxic 2.408*** 5.43 $101.61 2.787*** 5.78 $95.13 VOC 0.712** 2.25 $30.02 2.244*** 5.27 $76.59 LR index 0.26 0.37 Sunlight, tags, and flowering Price 0.061*** 14.95 n/a 0.085*** 14.58 n/a SunlightP 0.033 0.23 $0.54 0.187 1.1 0 $2.21 SunlightFD 0.466 *** 3.21 $7.69 0.501*** 3 .00 $8.27 Tags 0.580*** 4.83 $9.57 0.510*** 3.48 $8.42 Flowers 1.021*** 8.61 $16.85 0.663*** 4.68 $7.83 VOC 0.345*** 2.92 $5.70 1.723*** 9.98 $20.37 LR index 0.21 0.31 Sunlight, toxicity, and flowering Price 0.042*** 9.1 0 n/a 0.052*** 9.45 n/a SunlightP 0.244 1.63 $ 5.8 0 0.011 0.07 $ 0.21 SunlightFD 0.517*** 3.02 $12.32 0.489 *** 2.68 $9.50 Toxic 2.191*** 12.05 $52.17 2.243*** 12.07 $43.55 Flowers 1.022*** 7.11 $24.33 1.279*** 7.82 $24.83 VOC 1.045*** 7.03 $ 24.8 7 1.962*** 10.77 $38.10 LR index 0.29 0.37 Sunlight, tags, and t oxicity Price 0 .012** 2.48 n/a 0.061*** 11.54 n/a SunlightP 0.030 0.17 $2.51 0.034 0.25 $0.56 SunlightFD 0.706*** 3.5 $59.34 0.743*** 4.68 $12.26 Tags 0.407** 2.36 $34.19 0.848*** 5.64 $ 71.2 3 Toxic 2.231 *** 9.62 $187.48 2.178*** 13.17 $35.94 VOC 0.895*** 5.16 $75.23 2.167*** 11.97 $35.75 LR index 0.23 0.30 *,**, and *** indicate significance at the 90%, 95%, and 99% confidence levels, respectively

PAGE 89

89 Table 5 2. continued Parameter No VOC information VOC information Estimate T value WTP Estimate T value WTP Height, hardiness, and f lowering Price 0.034*** 7.16 n/a 0.037*** 6.93 n/a Hardy1 0.168 1.06 $5.03 0.199 1.12 $5.34 Hardy2 1.154*** 5.36 $ 34.1 9 0.712*** 3.23 $19.13 Height4 0.834*** 4.5 $24.88 0.610*** 3.07 $16.39 Height8 1.213*** 6.37 $36.21 1.036*** 5.04 $35.13 Flowers 1.398*** 8.73 $41.72 0.561*** 3.56 $15.07 VOC 0.419*** 2.68 $1.25 1.307 *** 6.83 $33.64 LR index 0.22 0.18 Hardiness, height, and t ags Price 0.030*** 4.87 n/a 0.061*** 5.85 n/a Hardy1 0.321* 1.68 $10.76 0.267 0.93 $4.42 Hardy2 1.676*** 5.33 $56.23 1.422*** 3.17 $23.51 Height4 0.488** 2.08 $16.39 1.037*** 3.02 $17.14 Height8 0.525** 2.36 $17.61 1.848*** 4.53 $30.54 Tags 0.423** 2.3 0 $14.20 0.513* 1.95 $8.48 VOC 0.957*** 4.7 0 $32.12 1.763*** 4.91 $29.14 LR index 0.17 0.36 Hardiness, height, and toxicity Price 0.031*** $ 5.09 n/a 0.015*** 3.34 n/a Hardy1 0.245 $ 1.03 7.81 0.072 0.43 $4.81 Hardy2 1.125*** $ 3.5 0 35.82 1.015*** 4.38 $68.09 Height4 0.082 $ 0.35 2.62 0.416** 2.18 $27.93 Height8 1.273*** $ 4.04 40.53 0.740*** 3.97 $49.67 Toxic 1.659*** $ 5.98 52.82 1.718*** 8.6 0 $115.28 VOC 1.197*** $ 4.79 38.11 1.679*** 8.69 $112.71 LR index 0.33 0.27 Sunlight, height, and flowering Price 0.022*** 9.46 n/a 0.034*** 10.66 n/a SunlightP 0.116 1.34 $5.38 0.029 0.26 $0.84 SunlightFD 0.198** 2 .00 $9.15 0.220* 1.74 $6.44 Height4 1.001*** 11.02 $46.33 0.785*** 6.59 $23.02 Height8 1.419*** 13.86 $65.71 1.033*** 8.63 $30.29 Flowers 0.673*** 9.22 $31.17 0.691*** 7.49 $20.28 VOC 0.447*** 5.69 $20.70 1 .337*** 11.92 $39.23 LR index 0.14 0.16 *, **, and *** indicate significance at the 90 % 95 % and 99% confidence level s, respectively

PAGE 90

90 Table 5 2. continued Parameter No VOC information VOC information Estimate T value WTP Estimate T value WTP Sunlight, height, and tags Price 0.021*** 7.32 n/a 0.046*** 13.07 n/a SunlightP 0.067 0.6 0 $3.27 0.043 0.41 $0.92 SunlightFD 0.152 1.34 $7.40 0.658*** 4.94 $14.11 Height4 0.016 0.15 $0.80 0.691*** 5.65 $14.90 Height8 0.049 0.42 $2.38 1.153*** 9.52 $24.86 Tags 0.178* 1.83 $8.67 0.739*** 7.94 $15.93 VOC 0.474*** 4.75 $23.12 1.905*** 14.68 $41.06 LR index 0.08 0.22 Sunlight, h ardiness and f lowering Price 0.046*** 16.26 n/a 0.058*** 17.26 n/a SunlightP 0.150* 1.74 $3.24 0.098 1.11 $17.01 SunlightFD 0.698*** 6.64 $15.06 0.639*** 5.73 $11.05 Hardy1 0.425*** 4.79 $9.18 0 .251*** 2.66 $4.34 Hardy2 1.616*** 13.6 0 $34.90 1.462*** 11.49 $25.27 Flowers 1.399*** 15.33 $30.22 0.851*** 9.52 $14.72 VOC 0.720*** 8.77 $15.55 1.753*** 17.91 $30.33 LR index 0.24 0.28 Sunlight, h ardiness, and t ags Price 0.050*** 16.13 n/a 0.058*** 17.37 n/a SunlightP 0.133 1.47 $2.64 0.248*** 2.72 $4.31 SunlightFD 0.461*** 4.22 $9.18 1.009*** 8.65 $17.55 Hardy1 0.286*** 3.1 0 $5.69 0.273*** 2.8 0 $4.74 Hardy2 1.823*** 12.72 $36.32 1.758*** 12.2 $30.58 Tags 0.711*** 8.04 $14.15 0.447*** 5.01 $7.78 VOC 1.242*** 13.85 $24.73 1.939*** 18.92 $33.72 LR index 0.25 0.31 Sunlight, hardiness, and toxicity Price 0.024*** 6.67 n/a 0.036*** 11.95 n/a SunlightP 0.049 0.37 $2.07 0.026 0.25 $0.73 SunlightFD 1.128*** 6.81 $ 47.3 9 0.894*** 6.98 $ 24.76 Hardy1 0.241* 1.74 $ 10. 11 0.060 0.51 $1.66 Hardy2 1.246*** 6.98 $52.37 1.055*** 7.92 $ 29.2 2 Toxic 2.437*** 13.14 $ 102.39 2.169*** 17.19 $ 60.0 9 VOC 1.196*** 9.77 $50.25 2.028*** 18.08 $56.17 LR index 0.35 0.37 *, **, and *** indicate significance at the 90 % 95 % and 99% confidence level s, respectively

PAGE 91

91 Table 5 2. continued Parameter No VOC information VOC information Estimate T value WTP Estimate T value WTP Sunlight, h eight, and t oxicity Price 0.018*** 5.4 0 n/a 0.031*** 8.69 n/a SunlightP 0.164 1.2 0 $8.90 0.117 0.88 $3.85 SunlightFD 0.004 0.03 $0.23 0.053 0.35 $1.75 Height4 0.586*** 4.23 $31.86 0.396*** 2.95 $12.98 Height8 0.962 *** 6.41 $52.28 1.139*** 7.33 $37.33 Toxic 1.184*** 8.92 $64.37 1.589*** 11.21 $52.10 VOC 1.142*** 8.6 0 $62.08 1.558*** 11.22 $51.07 LR index 0.17 0.25 Sunlight, h ardiness, and h eight Price 0.037 *** 11.88 n/a 0.035*** 14.06 n/a SunlightP 0.103 1.04 2.77 0.041 0.41 11.91 SunlightFD 0.687*** 6.26 18.58 0.734*** 5.81 21.27 Hardy1 0.273*** 3.02 7.37 0.233** 2.5 6.76 Hardy2 1.395*** 11.04 37.71 1.491 *** 11.72 43.22 Height4 0.869*** 8.75 23.49 0.217** 2.36 6.30 Height8 1.760*** 13.8 0 47.56 1.114 *** 9.91 32.29 VOC 0.516*** 6.25 15.17 1.566*** 17.3 45.38 LR index 0.22 0.25 *, **, and *** indicate significance at the 90 % 95 % and 99% confidence level s, respectively

PAGE 92

92 T able 5 3 Significance and sign of attribute levels Attribute Price Flowers Tags Toxic Height4 Height8 Hardy1 Hardy2 SunlightP SunlightFD VOC Flowering, tags, and toxicity (A) NS NS N/A N/A N/A N/A N/A N/A + Flowering, tags, and toxicity (B) + NS N/A N/A N/A N/A N/A N/A + Flowering, tags, and hardiness (A) + + N/A N/A N/A N/A N/A + Flowering, tags, and hardiness (B) + NS N/A N/A N/A NS N/A N/A + Hardiness, toxicity, and flowering (A) + N/A N/A N/A NS N/A N/A + Hardiness, toxicity, and flowering (B) NS N/A NS N/A N/A NS N/A N/A NS Hardiness, tags, and toxicity (A) N/A + N/A N/A NS N/A N/A + Hardiness, tags, and toxicity (B) N/A + N/A N/A N/A N/A + Height, tags, and flowering (A) + NS N/A N/A N/A N/A N/A + Height, tags, and flowering (B) + NS NS N/A N/A N/A N/A + Height, toxicity, and flowering (A) NS + N/A N/A N/A N/A N/A + Height, toxicity, and flowering (B) + N/A NS N/A N/A N/A N/A + A is without VOC information; B is with information

PAGE 93

93 Table 5 3. continued Attribute Price Flowers Tags Toxic Height4 Height8 Hardy1 Hardy2 SunlightP SunlightFD VOC Height, tags, and toxicity (A) N/A + NS N/A N/A N/A N/A + Height, tags, and toxicity (B) N/A NS N/A N/A N/A N/A + Sunlight, tags, and flowering (A) + + N/A N/A N/A N/A N/A NS + Sunlight, tags, and flowering (B) + + N/A N/A N/A N/A N/A NS + Sunlight, tags, and toxicity (A) N/A + N/A N/A N/A N/A NS + Sunlight, tags, and toxicity (B) N/A + N/A N/A N/A N/A NS + Height, hardiness, and flowering (A) + N/A N/A NS N/A N/A + Height, hardiness, and flowering (B) + N/A N/A NS N/A N/A + Hardiness, height, and tags (A) N/A + N/A N/A N/A + Hardiness, height, and tags (B) N/A + N/A NS N/A N/A + Hardiness, height, and toxicity (A) N/A N/A NS NS N/A N/A + Hardiness, height, and toxicity (B) N/A N/A NS N/A N/A + Sunlight, height, and flowers (A) + N/A N/A N/A N/A NS + Sunlight, height, and flowers (B) + N/A N/A N/A N/A NS + A is without VOC information; B is with information

PAGE 94

94 Table 5 3 continued Attribute Price Flowers Tags Toxic Height4 Height8 Hardy1 Hardy2 SunlightP SunlightFD VOC Sunlight, height, and tags (A) N/A N/A NS NS N/A N/A NS Sunlight, height, and tags (B) N/A + N/A N/A N/A NS + Sunlight, hardiness, and flowering (A) + N/A N/A N/A N/A + Sunlight, hardiness, and flowering (B) + N/A N/A N/A N/A NS + Sunlight, hardiness, and tags (A) N/A + N/A N/A N/A NS + Sunlight, hardiness, and tags (B) N/A + N/A N/A N/A + Sunlight, hardiness, and toxicity (A) N/A N/A N/A N/A + NS + Sunlight, hardiness, and toxicity (B) N/A N/A N/A N/A NS NS + Sunlight, height, and toxicity (B) N/A N/A N/A N/A NS NS + Sunlight, height, and toxicity (B) N/A N/A N/A N/A NS NS + Sunlight, hardiness, and height (A) N/A N/A N/A NS + Sunlight, hardiness, and height (B) N/A N/A N/A NS + Fixed Attributes (A) + N/A N/A N/A N/A NS + Fixed Attributes (B) + N/A N/A N/A N/A NS + A is without VOC information; B is with information

PAGE 95

95 *** indicates significance at the 99% confidence level Table 5 4 Parameter estimates and WTP for the fixed attribute model Parameter No VOC information VOC information Estimate T value WTP Estimate T value WTP Price 0.039 0*** 30.58 n/a 0.0498 *** 31.29 n/a SunlightP 0.0401 0.98 $1.03 0.0230 0.53 $0.46 SunlightFD 0.3453 *** 7.29 $ 8.85 0.4089 *** 7.59 $ 8.21 Hardy1 0.3092 *** 7.47 $ 7.93 0.2821 *** 6.01 $ 5.66 Hardy2 1.363 0*** 24.4 $ 34.95 1.2412 *** 20.97 $ 24.92 Flowers 0.6297*** 16.29 $16.15 0.4705*** 11.34 $9.45 VOC 0.9199 *** 23.76 $ 23.59 2.0438 *** 40.73 $ 41.04

PAGE 96

96 Table 5 5. Willingness to pay ranges for each attribute Attribute Attribute selection Fixed attributes No VOC information VOC information No VOC information VOC information Flowering $3.24 to $41.72 $7.83 to $32.16 $16.15 $9.45 Tags $8.67 to $71.23 $7.09 to $77.23 n/a n/a Toxicity $187.48 to $22.00 $115.28 to $35.94 n/a n/a Hardiness (needs some care) 12.15 to $10.11 $6.76 to $4.34 $7.93 $5.66 Hardiness (needs a lot of care) $56.23 to $52.36 $68.09 to $15.16 $34.95 $24.92 Height (4 to 8 feet) $87.14 to $2.62 $6.30 to $17.14 n/a n/a Height (8 to 12 feet) $148.50 to $2.38 $65.02 to $30.54 n/a n/a Sunlight (Partial) $3.24 $4.31 NS NS Sunlight (Full/Direct) 59.34 to $7.40 $24.76 to $6.44 $8.85 $8.21 VOC $23.12 to $75.23 $16.45 to $112.71 $23.59 $41.04

PAGE 97

97 Table 5 6. Willingness to pay weighted m eans Attribute No VOC information VOC information Hardy1 $ 2.00 $ 2.17 Hardy2 $ 11.31 $ 19.87 SunlightP $ 3.24 $ 4.31 SunlightFD $ 13.93 $ 11.65 Height4 $ 10.86 $ 5.70 Height8 $ 17.66 $ 12.93 Tags $ 2.80 $ 21.41 Flowers $ 11.52 $ 6.79 Toxic $ 24.29 $ 14.77 VOC $ 20.61 $ 39.01

PAGE 98

98 CHAPTER 6 CONCLUSION AND FUTURE RESEARCH The first objective of this research was to determine if the distribution of information on an indoor plants ability to remove indoor air pollution had an effect on consumer preferences and buying behavior The second objective was to determine if allowing p articipants in choice experiment to select the attributes most important to them would result in different willingness to pay estimates than when the investigator sele cts the attributes To achieve these objectives a choicebased conjoint, or CBC, in the form of a survey was emailed to over 2,200 participants. This CBC permitted approximately 1,500 participants to select attributes of indoor plants that they preferred. In a ddition, approximately half of all participants were provided with information on the ability of some indoor plants to remove indoor air pollution, commonly referred to as volatile organic compounds/chemicals, or VOCs. By comparing the results from the CBC between the half of the participants that received information on the ability of indoor plants to remove VOCs from indoor air with those that did not receive information, it was determined that information did have an effe ct on participants preferences. This impact was shown, not only through their preferences for the attribute of VOC removal but also for the other attributes. For the surveys allowing attribute selection a weighted average of willingness to pay for each attribute was created from the various combinations of attributes selected. The weighted average WTP for VOC re moval was $20.61 when information about VOCs was not given; however, when this information was given this increased to $39.01. This was an $18.40 increase in WTP or a nearly 90% increase. In the fixed attribute survey WTP with no information for VOC re moval was $23.59 and increased to $41.04 when information was given (a 74% increase).

PAGE 99

99 In addition to information on VOCs increasing WTP for the VOC attribute, WTP for other attributes changed. For the surveys allowing attribute selection the weighted m ean WTP increased for several of the other attributes including full/direct sunlight, both levels of height, tags, and the toxicity of the indoor plant The weight ed average WTP for full/direct sunlight increased from $13.93 (without information) to $11.65 (with information) a 16% increase. For a plant that would grow to 4 to 8 feet at maturity the weighted mean WTP increased 48%, from $10.86 without information to $5.70 with information. A slightly smaller increase (27%) was seen for plants that would grow to 8 to 12 feet at maturity from a weighted mean WTP of $17.66 without information to $12.93 with information. The largest percent change was for the attribute tag weighted average WTP for the plant having a tag increased from $2.80 when no information was provided to $21.41 when information was provided (a 665% increase) If a plant was toxic the weighted mean WTP was $24.29 if information was not given and increased to $14.77 when information was given (a 39% increase) A few of the at tributes experienced decreases in their weighted mean WTP including both levels of hardiness, partial sunlight, and the ability of the plant to flower. An indoor plant requiring some care had a weighted mean WTP of $2.00 when information was not provided which decreased by 9% to $2.17 when information was provided. If the plant required a lot of care, a 76% decrease in the weighted mean WTP occurred, from $11.31 when information was given to $19.87 when information was not given. The weighted mean WT P for the plants ability to flower also decreased, from $11.52 when information was not provided to $6.79 when it was provided (41%)

PAGE 100

100 Examining these changes in WTP shows that information does have an effect on consumer preferences and behaviors. The information about VOCs and the ability of some indoor plants ability to remove them not only has an impact on consumer preferences for this ability but also for the other attributes a plant may have. After being informed about this ability consumers are wi lling to pay more (or less) for different attributes than before they were given the information. For example, there was a large increase in the weighted mean WTP for tags clearly identifying the plant when information was provided. Consumers may want to ensure that the plants that they are purchasing specifically to remove VOCs are correctly identified. In terms of the second objective allowing participants to select the attributes that they prefer in a product can result in more information about consumer preferences and more accurate WTP estimates. In a traditional CBC there is a fixed set attributes from which participants are forced to choose a product containing some level of all the attributes. Some participants may not want some of these attributes or if they do want the attributes they may not prefer them as much as other possible attributes. Therefore, WTP may not be estimated accurately. Comparing the WTP for the attributes selected by participants that are the same attributes included in the fixed set further illustrates this point. In the fixed attribute survey, all participants participated in a choice experiment where attributes for indoor plants did not change. These attributes included price, VOC removal, sunlight, hardiness, and flowering. The differences in WTP for full/direc t sunlight for participants who selected attributes and those in the fixed survey attributes highlight the difference between the two methods. For attribute selection, without information the WTP for full/direct sunlight is $15.06 compared to a plant tha t requires little/indirect sunlight. With

PAGE 101

101 information this WTP increased to $11.05. In contrast, the WTP for full/direct sunlight in the fixed set was $8.85 when no information was provided and increased slightly to $8.21 when information was provided. Additionally, the impact of information differs between the two types of surveys. With attribute selection, a decrease of 27% is seen in the WTP when information is provided. This decrease is only 7% with fixed attributes. The WTP for an indoor plants ability to flower also demonstrates the diffe rence between the approaches. With attribute selection and no information on VOCs the WTP for a flowering plant was $30.22. This decreased to $14.72 when information about VOCs was given (suggesting VOCs beca me relatively more important and flowering became less important as an attribute) For the participants who did not select their top attributes the WTP for a flowering plant was $16.15 with no information on VOCs and decreased to $9.45 when information was given. Both showed decreases, though the values and the amounts differ. With attribute selection a flowering plant was valued at a larger amount and decreased 51% when information was given. Without attribute selection, the decrease was 41% but the initial value of a flowering plant was 47% lower. Though it is interesting to compare the mean willingness to pay, it is also interesting to consider the ranges of WTP estimates derived from the survey with attribute selection. This demonstrates the range of impact of heterogeneous preferences of consumers. With attribute selection WTP for the ability of the plant to flower ranges from $3.24 to $41.72 without information and from $7.83 to $32.16 with information. The WTP for flowering when attributes are fixed do es fall within these ranges; however, these ranges suggest consumers differ in how they value this attribute. Also, the weighted mean WTP for flowering when attributes are

PAGE 102

102 selected is $11.52 without i nformation and $6.79 with information. The WTP for flowering in the fixed set are $16.15 with information and $9.45 without information. This indicates that when consumers are provided with a fixed set of attributes their WTP for some attributes may be i nflated. The same issue occurs for full/direct sunlight. With attribute selection the range of WTP for this attribute was $59.34 to $7.40 when information was not provided and $24.76 to $6.44 when information was provided. In the case of fixed attri butes WTP was $8.85 when information was not given and $8.21 when the information was given. The WTP of participants who were given a fixed set of attributes may not have been their actual WTP as they were forced to make choices for a set of attributes t hat they may not actually prefer in an indoor plant. This conclusion can also be demonstrated by the actual choices of those participants included in the surveys with attribute selection. Height of the plant at maturity, the care required of the plant, a nd the sunlight needed were selected most frequently by participants, both when information was provided and when it was not. The attributes of which composed the fixed set, sunlight needed, care required, and the plants ability to flower, were chosen le ss frequently. Floridas floriculture industry can be positively impacted by the results of this research. As shown in Chapter 1 Florida has been especially affected by the recent recession, both in its housing market and floriculture sales. Though ho using prices are still below 2006 levels, the number of home sales has increased over the last four years. The increase in housing sales may indicate that consumers have more disposable income. This income can also be used to purchase indoor plants, espe cially if consumers are informed of the benefits of specific plants which remove indoor air pollution. If consumers are given information about how their homes can

PAGE 103

103 contain VOCs, the possible impacts of VOCs on health, and the ability of some indoor plants to remove them floriculture sales may increase. Consumers may see indoor plants as more than decoration; they may view these plants as necessities.. Participants in the attribute selection survey without information were willing to pay $20.61 for VOC re moval and participants who were included in the fixed attribute survey without information were willing to pay $23.59. This indicates with very little marketing on the part of the floriculture industry, consumers would respond positively to plants with information that indicated they reduce VOCs. However, with marketing and education, the willingness to pay increases. When information w as provided to participants about VOCs and indoor plants ability to remove them the weighted mean WTP (in the survey with attribute selection) for VOC removal increased to $39.01. For t he fixed attribute s s urvey WTP for VOC removal increased to $41.04. Information clearly had an effect on how participants viewed this attribute. It should be noted that even those participants who did not receive this information WTP for VOC removal was always positive and greater than WTP for some other attributes. This result indicates that consumers also prefer an indoor plant that will perform a function even if the y are not completely informed about that function. Informing participants about VOCs also had an impact on how they viewed other attributes. For the survey with attribute selection there were clear differences in the order of preferences for the attribut es without and with VOC information These differences can be shown through weighted mean WTP in Table 61. Without information flowering was the most preferred attribute as it had the greatest weighted mean WTP and toxicity was the least preferred attr ibute as it had the smallest weighted mean WTP. When this information was provided

PAGE 104

104 weighted mean WTP was greatest for tags, indicating this attribute was most preferred, and weighted mean WTP for needs a lot of care was the smallest, indicating this attrib ute was the least preferred. Tags experienced an increase in its weighted mean WTP while there was a decrease in WTP for flowering. Toxicity, though not prefe rred among participants, was no longer the least preferred attribute when information was provided. Instead, if the plant required a lot of care participants preferred this attribute the least if they were given information Informing consumers may lead to them valuing attributes differentl y than if they were not given information. By distributing this information growers can then determine which attributes are most important to co nsumers and provide plants to consumers that contain these attributes. For example, a consumer who has been gi ven information on VOCS may decide purchase an indoor plant that will remove them. However, this consumer will be more likely to purchase and be willing to pay more for a plant that removes VOCs and has a tag clearly indentifying it than a plant that remo ves VOCs and has the ability to flower. Examining how consumers react to this information can help Floridas floriculture industry market these plants as being necessary to have in homes which may lead to increase in floriculture sales. Future research, if feasible, could focus on a larger sample size. This study had over 2200 participants but a bigger sample size would yield more accurate results. Different and/or more attributes and levels could be used as well if the information does not overwhelm pa rticipants. Also, the multinomial logit should be considered in addition to the conditional logit. The multinomial logit analyzes participants choices based on the characteristics of the participants. The results may show which specific attributes of participants determine why they purchase certain indoor plants. Another issue to be explored is hypothetical bias. Participants in

PAGE 105

105 these surveys did not actually purchase indoor plants so they were not spending their own money and may have overstated how much they were willing to pay. Lusk and Schroeder (2004) investigated this hypothetical bias which can occur when conducting surveys and found that when not making actual purchases participants tended to overstate their WTP. Using another method, such as auctions, can help reduce this bias.

PAGE 106

106 Table 6 1. Order of preferences for attributes (descending) Order Without VOC information With VOC information 1 Flowering Tags 2 Tags Flowering 3 Needs some care Needs some care 4 Requires partial sunlight Requires partial sunlight 5 Grows to 4 to 8 feet tall Grows to 4 to 8 feet tall 6 Needs a lot of care Requires full/direct sunlight 7 Requires full/direct sunlight Grows to 8 to 12 feet tall 8 Grows to 8 to 12 feet tall Toxicity 9 Toxicity Needs a lot of care

PAGE 107

107 REFERENCE LIST Absolute Air Cleaners, Air Purifiers, & Allergy Products 2010. Air Cleaners & Air Purifiers. http://www.aircleaners.com/items.php?CID=104 Consumer Reports 2007. Proposed settlement in Sharper Image air purifiers case. If you own an Ionic Breeze model, you might qualify for merchandise credit. ConsumerReports.org. Consumers Union of U.S., Inc., February 2007. http://www.consumerreports.org/cro/homegarden/news/2007/02/settlement in sharper image ionic breezeclass action 207/overview/0702_sharper image.htm Green, P.E., A.M. Krieger, and M.K. Agarwal. 1991. Adaptive Conjoint Analysis: Some Caveats and Suggestions. Journal of Marketing Research. 28(2): 215222. Green, P.E. and V. Srinivasan. 1978. Conjoint Analysis in Consumer Researcher: Issues and Outlook. The Journal of Consumer Research. 5(2): 103123. Green, P.E. and V. Srinivasan. 1990. Conjoint Analysis i n Marketing: New Developments with Implications for Research and Practice. Journal of Marketing. 54(4): 319. Gustafsson, A., A. Herrmann, and F. Huber Conjoi nt Analysis as an Instrument of Market Research Practice. Conjoint Measurement : Methods and Applications. Ed. Gustafsson, A., A. Herrmann, and F. Huber Berlin, New York: Springer, 2001. 546. Print. Hall, C.R., B.L. Campbell, B.K. Behe, C. Yue, R.G. Lopez, and J. Dennis. 2010. The Appeal of Biodegradable Packaging to Floral Customers. HortScience. 45(4): 583591. Hodges, A., M. Palma, and C. Hall. 2010. Trade Flows and Marketing Practices within the U.S. Nursery Industry, 2008. Southern Cooperative Series S 1021 Multi state Regional Project Housing and Urban Development (HUD). 2007. U.S. Housing Market Conditions 4th Quarter 2007. http://www.huduser.org/periodicals/ushmc/winter07_1/index.html Housing and Urban Development (HUD). 2008. U.S. Housing Market Conditions 4th Quarter 2008. http://www.huduser.org/portal/periodicals/ushmc/winter08/index.html Housing and Urban Development HUD. 2009. U.S. Housing Market Conditions 4th Quarter 2009. http://www.huduser.org/portal/periodicals/ushmc/winter09/index.html Housing and Urban Development HUD. 2010. U.S. Housing Market Conditions 4th Quarter 2010. http://www.huduser.org/portal/periodicals/ushmc/winter10/index.html

PAGE 108

108 Indoor Air Quality. 2010a. An Introduction to Indoor Air Quality Environmental Protection Agency. http://www.epa.gov/iaq/voc.html Indoor Air Quality. 2010b. Indoor Air Facts No. 4 (revised) Sick Building Syndrome Environmental Protect ion Agency. http://www.epa.gov/iaq/pubs/sbs.html#Causes%20of%20Sick%20Building%20Syndrom e Indoor Air Quality. 2010c. Ozone Generators that are Sold as Air Cleaners. Environmental Protection Agency. http://www.epa.gov/iaq/pubs/ozonegen.html Kobayashi, K.D., A.J. Kaufman, J.Griffis, and J. McConnell. 2007. Using Housplants to Clean Indoor Air. Cooperative Extension Service, College of Tropical Agriculuture and Human Resources, University of Hawaii at Manoa. Liu, Y., Y. Mu, Y. Zhu, H. Ding, and N.C. Arens 2007. Which Ornamental Plant Species Effectively Remove Benzene from Indoor Air ? Atmospheric Environment 41: 650654. Long, J.S. 1997. Regression Models for Categorical and Limited Dependent Variables Advanced Quantitative Techniques in the Social Sciences Series. Thousand Oaks, CA: Sage Publications, Inc. Lusk, J.L. a nd T.C. Schroeder. (2004). Are Choice Experiments Incentive Compatible? A Test with Quality Differentiated Beef S teak American Journal of Agricultural Economics. 86(2): 467482. Lusk, J.L and F.B. Norwood. 2005. Effect of Experimental Design on Choice Based Conjoint Valuation Estimates. American Journal of Agricultural Economics. 87(3): 771785. Mcfadden, D. 1980. Econometric Modeling for Probabilistic Choice Among Products. The Journal of Business. 53(3): S13S29. Maddala, G.S. 1983. Limited Dependent Variables and Qualitative Variables in Econometrics. Cambridge: University Press. Mason, S.C., T.W. Starman, R.D. Lineberger, and B.K. Behe. 2008. Consumer Prefernces for Price, Color Harmony, and Care Information of Container Gardens. HortScience. 43(2): 380384. Matsumoto, H. and M.Yamaguchi. (2007) Experimental Study on the Effect of Foliage Plants on Removing Indoor Air Cont aminants. Proceedings of Clima 2007 Well Being Indoors.

PAGE 109

109 National Agricultural Statistics Service (NASS). 2002. Foliage, Floriculture, and Cut Greens. http://www.nass.usda.gov/Statistics_by_State/Florida/Publications/Horticul ture/f&fcg02.pdf National Agricultural Statistics Service (NASS). 2003. Foliage, Floriculture, and Cut Greens. http://www.nass.usda.gov/Statistics_by_State/Florida/Publications/Horticul ture/f&fcg03.pdf National Agricultural Statistics Service (NASS). 2005. Foliage, Floriculture, and Cut Greens. http://www.nass.usda.gov/Statistics_by_State/Florida/Publications/Horticul ture/f&fcg05.pdf National Agricultural Statistics Service (NASS). 2007. Foliag e, Floriculture, and Cut Greens. http://www.nass.usda.gov/Statistics_by_State/Florida/Publications/Horticul ture/f&fcg07.pdf National Agricultura l Statistics Service (NASS). 2009. Foliage, Floriculture, and Cut Greens. http://www.nass.usda.gov/Statistics_by_State/Florida/Publications/Hor ticul ture/f&fcg09.pdf National Agricultural Statistics Service (NASS). 2011. Foliage, Floriculture, and Cut Greens. http://www.nass.usda.gov/Statistics_by_State/Florida/Publications/Horti culture/f&fcg11.pdf Orwell, R.L., R.L. Wood J. Tarran F. Torpy, and M.D. Burchett. 2004. Removal of Benzene by the Indoor Plant/Substrate Microcosm and Implications for Air Qual ity. Water, Air, and Soil Pollution. 157: 193207. Oyabu, T., A. Sawada, T. Onodera, K. Takenaka, and B. Wolverton. 2003. Characteristics of Potted Plants for Removing Offensive Odors. Sensors and Actuators. 89: 131136. Papinchak H.L., E.J. Holcomb, T.O. Best, and D.R. Decoteau. 2009. Effectiveness of Houseplants in Reducing the Indoor Air Pollutant Ozone HortTechnology. 19(2): 286290. Ryan, M. and J. Hughes. 1997. Using Conjoint Analysis to Assess Women's P references for Miscarriage Management Health Economics. 6(3): 261273. SAS Institute Inc. 1993. SAS Technical Report R 109, Conjoint Analysis Examples Cary, NC: SAS Intstiute Inc., 1993.85 pp. SAS Institute Inc. 2010. SAS/ETS 9.22 Users Guide. Cary, NC: SAS Institute Inc.. http://support.sas.com/documentation/cdl/en/etsug/63348/PDF/default/etsug.pdf

PAGE 110

110 Sawtooth Software. 2008. CBC v6.0 Technical Paper. Technical Paper Series. Sawtooth Software, Inc. http://www.sawtoothsoftware.com/download/techpap/cbctech.pdf Vriens, M. Conjoint Analysis in Marketing. Diss. Rijksuniversiteit te Groningen, 1995. Wang, S., H.M. Ang, and M.O. Tade. 2007. Volatile Organic Compounds in Indoor Environment and Photocatalytic Oxidation: State of the Art. Environmental International. 33: 694705. Wittink, D.R. and T. Bergestuen. Forecasting with Conjoint Analysis. Principles of Forecasting: A Handbook for Researchers and Practitioners. Ed. J. Scott Armstrong 2001. New York, NY: Spring Science + Business Media, Inc., 2001. 147 149. Print. Wittink, D.R., L. Krishnamurthi, and D.J. Reibstein. 1989. The Effect of Differences in the Number of A ttribute Levels on Conjoint Results. Marketing Letters. 1(2): 113123. Wolverton, B.C., A.J. Johnson, and K. Bounds. 1989. Interior Landscape Plants for Indoor Air Pollution Abatement. Final Report for the National Aeronautics and Space Administration. Wolverton, B.C. and J.D. Wolverton. 1993. Plants and Soil Microorganisms: Removal of Formaldehyde, Xylene, and Am monia from the Indoor Environment. Journal of the Mississippi Academy of Sciences. 38: 1115. Wood, R.A., M.D. Burchett, R. Alquezar, R.L. Orwell, J. Tarran, and R. Torpy. 2006. The Potted Plant Microcosm Substantially Reduces Indoor air VOC Pollut ion: I. Office Field Study. Water, Air, and Soil Pollution. 175: 163 180. Woolston, C. Ionic purifiers' dirty secret Las Angeles Times. n.p., 21 April 2008. http://articles.latimes.com/2008/apr/21/health/he skeptic21 Yue, C., and B.K. Behe. (2008). Estimating U.S. Consumers Choice of Floral Retail Outlets. HortScience. 43(3): 764 769.

PAGE 111

111 BIOGRAPHICAL SKETCH Alexis Solano is originally from Newark, DE. She completed her bachelors d egree in economics in 2004 and her masters degree in agricultural and resource e conomics in 2008, both from the University of Delaw are. She earned her PhD in food and resource e conomics from t he University of Fl orida in 2012.