1 T RAINING DOMESTIC DOGS (CANIS LUPUS FAMILIARIS) ON A NOVEL ODOR DETECTION TASK IN DISCRETE TRIALS By NATHANIEL JAMES HALL A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2012
2 2012 Nathaniel James Hall
3 To my loving wife and f amily
4 ACKNOWLEDGMENTS We would like to thank all of the owners that permitted our repeated testing of their dogs. We thank Camp Marlin Doggie Day Care for all of their help in recruiting and testing dogs. We would also like to thank Andrew White for his assistance with this pro ject.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF FIGURES ................................ ................................ ................................ .......... 6 ABSTRACT ................................ ................................ ................................ ..................... 7 CHAPTER 1 INTRODUCTI ON ................................ ................................ ................................ ...... 9 2 EXPERIMENT 1 ................................ ................................ ................................ ..... 13 Methods ................................ ................................ ................................ .................. 13 Subjects ................................ ................................ ................................ ............ 13 Materials ................................ ................................ ................................ ........... 13 Procedure ................................ ................................ ................................ ......... 14 Session 2 ................................ ................................ ................................ ................ 18 Statistical Analyses ................................ ................................ ................................ 18 Results and Discussion ................................ ................................ ........................... 19 3 EXPERIMENT 2 ................................ ................................ ................................ ..... 28 Methods ................................ ................................ ................................ .................. 28 Subjects ................................ ................................ ................................ ............ 28 Materials ................................ ................................ ................................ ........... 28 Procedures ................................ ................................ ................................ ....... 29 Statistical Analyses ................................ ................................ ................................ 31 Results and Discussion ................................ ................................ ........................... 32 4 GENE RAL DISCUSSION ................................ ................................ ....................... 40 LIST OF REFERENCES ................................ ................................ ............................... 44 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 46
6 LIST OF FIGURES Figure page 2 1 Layout of Experiment 1 showing experimental bins and a do g responding. ...... 24 2 2 Dog performance on Session 1 of Experiment 1.. ................................ .............. 25 2 3 The effects of food deprivation on performan ce in Session 1 ............................ 26 2 4 Dog performance on Session 2 of Experiment 1. ................................ ............... 27 3 1 Average performance across all five sessions of Experiment 2. ......................... 36 3 2 Acquisition of the odor discrimination for experimenter delivered food trials in Experiment 2.. ................................ ................................ ................................ .... 37 3 3 Cumulative performance across all trials in Experiment 2 ................................ 38 3 4 Comparisons of trial types and tasks in Experiment 2. ................................ ....... 39
7 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for th e Degree of Master of Science T RAINING DOMESTIC DOGS (CANIS LUPUS FAMILIARIS) ON A NOVEL ODOR DETECTION TASK IN DISCRETE TRIALS By Nathaniel James Hall May 2012 Chair: Clive D.L. Wynne Major: Psychology Dogs can be trained to reliably detect a wide variety of important odors. Little is known, however, about the rate at which dogs can learn to detect an odor, the variables that influence this rate, and how this rate may vary across individual dogs. In two experiments, we developed a discrete trials procedure to assess the rate at which previously untrained dogs acquire an odor detection task. In Experiment 1, 20 dogs were traine d to indicate the presence of a novel odor significantly above chance within 24 trials, with two dogs meeting an individual criterion for performing significantly above chance. Experiment 2 compared performance on a visual task (black vs. white) to an odor detection task and assessed whether placing accessible food directly in the target bin and inaccessible food in the non target bin enhanced discrimination training, compared to a procedure in which the experimenter delivered food shortly after responding. Dogs learned faster on the odor task when the experimenter delivered food, compared to when food was placed directly in the target bin. When the experimenter delivered food, dogs performed better on the odor task than the visual task. Performance on the visual task was unaffected by how food was delivered. Individual
8 differences were observed in the rate of acquisition of the odor task when the experimenter delivered food. This discrete trials procedure shows promise in the selection of dogs likely to per form well on an odor detection task and to evaluate experimental variables that may influence the rate dogs learn an odor detection.
9 CHAPTER 1 INTRODUCTION Dogs are accurate and reliable biosensors, making them a useful detector technology (Furton & Myers, 2001). Domestic dogs can detect a wide variety of odors including explosives (for a review see Goldblatt, Gazit & Terkel, 2009), narcotics (De an, 1972), tortoises (Cablk, Sagebiel, Heaton & Valentin, 2008), cows in estrus (Fischer Tenhagen, Wetterholm, Tenhagen & Heuwieser, 2011; Kiddy, Mitchell, Bolt & Hawk, 1978), prostate cancer in humans (Cornu, Cancel Tassin, Ondet, Giardet & Cussenot, 2011 ), bladder cancer (Willis et al., 2004), and numerous other volatile stimuli. Dogs can detect a target odor even in the presence of higher concentrations of extraneous odors (Waggoner et al., 1998), and can be trained to simultaneously detect multiple targ et odors (Williams & Johnston, 2002). In addition, dogs can detect odors in concentrations of tens of parts per billion for chemicals such as nitroglycerin (Johnston, 1999). at response to the presence of to a novel odor. Traditionally, dogs require extensive and intensive training to reach certification standards in odor detection. Sinn, G osling & Hilliard (2010) reported that the 341 st Training Squadron at Lackland Air Force Base, Texas, trains specially selected dogs for an average of 100 days (SD = 34 days) before deeming them ready for certification testing. Cornu et al. (2011) trained one dog for five days a week for 16 months befor e the dog accurately identified samples of urine from individuals with prostate cancer. These important odor discriminations require extended
10 periods of training, and make an analysis of the potential variables that influence the rate at which a dog learn s an odor detection, in general, difficult to isolate. Some researchers have utilized other, potentially more salient, odors, to train odor detection in dogs in relatively few trials. Williams and Johnston (2002) trained dogs to alert to ten different odo rs, one at a time, with the first discrimination requiring, on average, 28 trials. Williams and Johnston also found the number of trials required to reach criterion decreased as additional odor detections were trained. Fischer Tenhagen et al. (2011) traine d dogs to discriminate bovine estrus vaginal samples from diestrus samples within 52 reinforced trials. Importantly though, the dogs in both of these studies were not nave to odor detection. These dogs were first trained to alert on another and training data during the first discrimination, however, were not reported, as the training of the alerting response is confounded with the training of the odor detection. Thus, the rate at which nave dogs can be systematically trained to alert to a novel odor, for the first time has not previously been reported. Not all dogs that enter detection training programs successfully complete their training, making individual differe nces in acquisition an important area of interest for the effectiveness of these programs. Sinn, Gosling & Hilliard (2010) reported that 20.9% of dogs selected for the 341 st Training Squadron never met certification criterion for either odor detection or p atrol training. Similarly, the Transportation Security Administration (TSA) reported that approximately 50% of the puppies raised for odor detection are successfully trained (TSA, 2011). To maximize the percentage of dogs meeting certification, dogs are t ypically given a battery of tests aimed at identifying those most
11 likely to succeed as detector dogs. Sinn, Gosling & Hilliard evaluated the selection test used by the 341 st to static objec ts, thrown objects, different environments, and many other conditions. Sinn, Gosling, and Hilliard found that higher scores on the selection test increased the odds the dog would achieve certification as a patrol dog (a non odor detection dog) or a dual ce rtified dog (patrol and odor certified); however, higher scores did not increase the odds of a dog being certified in odor detection. Maejima et al. (2006) analyzed performing a principle component analysis, the seven behavioral traits were reduced to eased the odds, although marginally, of a successful outcome (odds ratio: 1.144; 95% CI: 1.085 1.206). performance on an odor detection task. A brief odor detection task in a contro lled likelihood to succeed as an odor detector than the presently deployed aptitude tests. The purpose of the present study was to develop a rapid systematic training procedu re for odor discrimination in dogs. In order to ensure the most objective record trials procedure, every presentation of the training stimuli is programmed as a trial and the trials procedure ensures that the rate at
12 which each animal acquires the task is assessed, which in turn, makes possible the identification of variables that influence performance. In Experiment 1, we investigated whether dogs could be trained to alert to a novel odor within 24 scheduled trials, and if we could identify variations in performance across dogs. In Experiment 2 we explored variations in the procedure of Experiment 1 as variables influencing performance and attempted to identify consistent high performing dogs to further identify individual differences in performance across dogs.
13 CHAPTER 2 EXPERIMENT 1 The purpose of Experiment 1 was to develop a discrete trials procedure to train dogs to alert to a novel odor. We sought to identify individual differences in performance and assess whether dogs can be trained to perform significantly above ch ance within 24 scheduled trials. Methods Subjects Twenty five pet do gs were selected for this study, of which, twenty completed Session 1 and sixteen completed Session 2. None of the subjects were working detector dogs or had any previous training to be o dor detecting dogs. All dogs were tested in a familiar environment. Materials Dogs were trained with discrete trials in a two choice procedure to root in anise scented pine shavings in Ste ri lite TM plastic bins (Sterilite Corporation, Townsend, MA) All tra ining, including the training of the alerting response, was done within scheduled trials of the experiment. Larg e dogs were trained to root in large bins (40 cm by 35 cm by 16.5 cm) whereas small dogs were trained to root in smaller (30 cm by 36 cm by 15 cm ) bins. The rooting response was chosen as the a lerting behavior because an observer could objectively score rooting, dogs were able to sniff both choices directly at the source before a response was made, and limited training for an alerting behavior w a s required (F igure 2 1 ). The bins were filled to a depth of approximately 8 cm with PetsPick TM pine shavings (American Wood Fibers, Columbia, MD) Pine shavings were placed in the
14 bins at least one hour prior to testing to allow their natural odor to dis sipate The target odor was anise extract. Anise extract was selected as the target as it was likely a novel odor to all household dogs, safe, readily available, and is utilized as a target odor by the National Association for Canine Scent Work (NACS, 2011). The target odor was prepared by placing 1 ml of McCormick TM (McCormick & Company, Inc., Sparks, MD) pure a nise extract on 100% cotton rounds using a measuring syringe. The scented cotton rounds were buried in the target containers approximately 2 .5 cm deep Before each trial, the two bins were placed at locations marked with masking tape 0 .25 m apart (see Figure 2 1 ) The subject was held at least 1 m back from the bins, and was released at the beginning of each trial. Before the dog was released, the experimenter steppe d at least 1 m away from the bins and looked straight down at the ground. An observer naive to which bin was correct stood 1 m back at the starting nse. After each trial, the naive observer would call the dog back. P rocedure Alert training : d ogs were first tested for motivation and were trained to in the pine For alert training trials, only one bin, the anise scented bin, was utilized. A treat wa s placed in an open tea ball, which was placed on top of the pine shavings and the anise scented cotton round. An open tea ball was used during alert training trials to keep the presence of the tea ball consistent with later food buried trials (described below). Two alert training trials were conducted where th e treat was visible and on top of the pine shavings. The dog was allowed to freely approach and consume the visible treat. If the dog did not consume the treat during these two trials, the experimenter would hand the treat to the dog. If the dog still did not consume the treat after two
15 attempts, the treat type was changed. Most dogs readily consumed commercial dog treats or cheese. I f, upon changing the treat, the dog still did not take any of the available food, testing was terminated for that dog. Only o ne dog failed to take any treats during this training. Dogs completing the first two trials were given three trials in which th e tea ball, the treat, and the a nise cotton round were buried in the pine shavings. Dogs were required to root to obtain the acce ssible food. Once the dog found ed the dog to eat the treat. All dogs completing the first two trials completed the following three trials. Food buried trials: f or food buried trials dogs were pr es ented with two bins; an anise scented bin with buried accessible food (S + ), and an identical bin without anise but with buried inaccessible food (S ). The purpose of food buried trials was to pair accessible food with the target odorant while strengthenin in the pine. Food was placed in an open tea ball in the S + bin (accessible), and food was placed in a closed tea ball in the S bin (inaccessible; Figure 2 1 ). The tea ball was utilized to bury identical pieces of food in the S + and S bins, making the anise odor, and not food odor, the only predictor of food. If food had been buried in only one bin, it would have been unclear if the dog was learning to detect food odor or the target odor Pilot studies also indicated that if a whole piece of food was buried in the S + bin, and an identical amount of food cut sufficiently small so as to be inaccessible in the pine was buried in the S bin, at least one dog could discriminate the difference in the food presentation Thus, the t ea ball was required to equate food odor between the S + and the S bins.
16 Before the start of each discrimination trial, the experimenter prepare d the S + bin by burying the anise scented cotton round and the open tea ball with a piece of food 2.5 cm deep in the pine shavings. The S bin was prepared by burying an unscented cotton round and a tea ball closed with a treat 2.5 cm deep in the pine shavings. When the dog was at the start location the experimenter placed the bins at the marked locations, steppe d 1 m back, and looked at the ground. Dogs were free to respond in any way. The naive observer watched the dog, and waited for the dog to root in either bin. Once experimenter to look up. If the experimenter observed the dog in the S + container, the accessible food (Figure 2 1 ). If the dog was rooting in the S bin, the experimenter picked up both bins, and began prepari ng for the next trial. If the dog had not made a choice after 1 minute both bins Experimenter delivered food trials : e xperimenter delivered food trials and food buried trials differed in that for the experimenter delivered food trials, no food was buried and the experimenter delivered the reinforcer after the dog made a choice. The S + bin contained an anise scented cotton round buried 2.5 cm in the pine, while the S bi n contained an unscented cotton round buried in the pine. Different sets of identical bins were utilized for food buried trials and experimenter delivered food trials to limit food cross contamination. The purpose of experimenter delivered food trials wa s to assess whether the dogs could also be trained to alert (i.e. root) in the bin with the target odor in the absence of any food odor that may prompt rooting.
17 Before each trial, the experimenter prepared the bins by burying the appropriate cotton roun d in the pine The experimenter placed the bins at the marked location, stepped 1 m back, and looked s traight at the ground. The naive observer watched the d in either container (Figure 2 1 ). The experimenter t hen looked up, to see the bin the dog had chosen. If the dog was rooting in the S + bin, the experimenter deliver a treat by hand. If the dog was rooting in the S bin, the experimenter picked up both bins and prepared for the next trial. If no response was made within 1 minute Control trials: control trials were conducted to test whether the dogs were utilizing unintentional cues in addition to or instead of the anise odor to identify the S + bin. The only difference between control trials and food buried trials was that neither bin contained an anise scented cotton round For the S + bin, the experimenter buried an open tea ball wit h a treat and an unscented cotton round 2.5 cm in the pine. For the S bin, the experimenter buried a closed tea ball with a treat inside and an unscented cotton round 2.5 cm in the pine. Thus, we expected an above chance performance on control trials if d ogs were discriminating between an open and closed tea ball, or if the at chanc e if the dogs were using only the anise odor to identify the S + bin. A total of six control trials were run for each dog, unless the dog chose correctly on five of the six control trials. If the dog chose correctly on five of six control trials, two to three additional trials were completed for that dog. Programmed trial order: d ogs were given 12 food buried trials 12 experimenter delivered food trials, and six control trials per session. These trials were divided into five
18 blocks of five trials (four training trials and one control). The initial block contained four food buried trials and one control trial. Food buried trials were subsequently decreased to two trials across the following three blocks and were faded out entirely for the last block. No experimenter delivered food trials were given in the first block, they were increased to two t rials per block in blocks 2, 3 and 4, and block 5 consisted of all experimenter delivered food trials. The trials were structured in different blocks to initially strengthen the rooting response with food buried trials, and to slowly fade in experimenter d elivered food trials where no food was buried. For all trial types, the location of the target (S + ) bin was pseudorandomly determined so that the S + bin was not at the same location for more than two trials in a row. If the dog made an incorrect choice and the previous four choices had also been to the same location a forced correction trial was run by repeating the same trial after the non target (S ) bin was picked up and made unavailable. If the dog made three incorrect choices in a row or two no choice s in a row two alert training trials with the food on top of the pine shavings were run to insure motivation. If the dog did not consume the food during both of these trials, testing was terminated for that dog. Testing was terminated for four dogs after failing to take freely available, visible, food. Session 2 Sixteen of the original twenty dogs were available to be retested for a second session. Dogs were re tested between 1 and 28 days after the first session (average of 11 days). All procedures were identical to those of Session 1. Statistical Analyses Performance on food buried and experimenter delivered food trials was analyzed both separately and in combination. One sample t tests were used to compare the
19 group performance on each trial type an d the overall average to chance. A two tailed binomial test was used to identify the criterion for an individual significantly above chance on the combined score (18 out of 24, 75%, p < .023) Paired t tests were used to compare the differences between food buried trials and experimenter delivered food moment correlation assess ed the correlation between S ession 1 and S ession 2. A linear regression assess ed the effect of the number of hours since the dog had last eaten before testing on performance. All statistical tests were run using R TM or Graphpad Prism TM Results and Discussion Performance varied across dogs, with only two of the twenty dogs meeting the individual criterion for performing significantly above chance in a single session (18 correct out of 24, 75%; Figure 2 1). Most dogs performed slightly above chance (chance is 50%); however, they did not meet the individual criterion. Individ ual performance on food buried trials ranged from 25% to 75% correct. Performances on experimenter delivered food trials ranged from 17% to 83% correct. Performance on control trials varied around chance. Overall, dogs showed a wide range of performances, with some achieving slightly higher levels of performances in experime nter delivered food trials ( Figure 2 2). buried and experimenter delivered food trials, was significantly a bove chance (Figure 2 2, one sample t test, t = 4.05, df = 19, p < .001). Performance was also significantly above chance when considering food buried trials alone (one sample t test, t = 3.22, df = 19, p < .01) or experimenter delivered food trials alone (on e sample t test, t = 2.98, df = 19, p < .01), and there was no statistical difference in performance between food buried
20 trials and experimenter delivered food trials (paired t test, t = .29, df = 19, p > .8). On control trials, performance remained at c hance (one sample t test, t = 1.7, df = 19, p > .05). Although at the individual level only two of twenty dogs performed significantly above chance, the average performance across dogs indicated that as a group, dogs identified the anise scented containe accuracy, however, never reached very high levels (average 60.7%), indicating that even as a group, the dogs were not selecting the anise scented container much more than expected by chance alone. The tim e since the last feeding (hereafter referred to as food deprivation) was estimated for each subject as the difference (to the nearest half hour) between the time the owner last reported feeding the dog and the time testing of the dog began. The ages of the dogs were owner reported and recorded to the nearest year. A liner regression model including the hours of food deprivation, age, and an interaction term revealed that the age of the dog had no significant effect on the total score ( p > .57), but the leve l of food deprivation had a significant negative effect ( p < .02: i.e. the more food depri ved the poorer the total score, Figure 2 3). No interaction between food deprivation and age was detected ( p > .79). A reduced model only including food deprivation a s a predictor variable showed that food deprivation had a significant .44 association, F(1,18) = 24.03, p < .001, with the total score ( R 2 = .57). Performance in Session 2 was highly similar to performance in Session 1; the average percent correct in the total score did not change from Session 1 to Session 2 (Session 1, 60.7% correct; Session 2, 60.7% correct). Performance on experimenter delivered f ood trials increased from a mean of 60% to a mean of 66%, whereas
21 performance on food buried trials decreased from a mean of 60% to 54% (Figure 2 4) The group performance on the experimenter delivered food trials and the combined total score was significa ntly above chance (one sample t test, t =4.21, df = 15, p < 05), however, when considering food buried trials alone, performance was at chance in Session 2 (one sample t test, t = 1.50, df = 15, p >.05). No statistically significant differences were observ ed between experimenter delivered food trials and food buried trials on a paired t test ( t = 1.78, df = 15, p > .05), however, the data suggest higher performances in experime nter delivered food trials ( Figure 2 4). Performance on control trials remained at chance. Unlike Session 1 a linear regression analysis revealed no statistically significant effect of food deprivation on performance in Session 2. A Pearson product moment correlation coefficient of .16 was observed for the correlation between Session 1 and Session 2; however, this was not significantly different from zero ( t = .6, df = 14, p < .56). Using this discrete trials procedure dogs were detecting a novel target odor significantly above chance within 24 trials. Prior to the scheduled trials, all dogs were nave to the task, and all training, including the training of the alerting response, occurred in programmed trials. This is an important difference from prior research in which the alerting response to an odor had been trained prior to expe rimental training. This procedure identified individual differences in performance across dogs: including two dogs that performed significantly above chance in a single session. Thus, the procedure may be useful in selecting dogs that learn an odor detect ion task rapidly.
22 In addition, the procedure identified that food deprivation may have a significant negative effect on performance; however, this effect was not replicated in Session 2. The average performance across Sessions 1 and 2 remained stable, ho wever, individual performance in Session 1 did not correlate with performance in Session 2. While the performance of some dogs increased, the performance of other dogs decreased equivalently. The average 11 day gap between Session 1 and Session 2 may have contributed to the lack of a correlation. Repeated testing less than 11 days apart may reveal a stronger test retest relationship. No significant differences were observed between food buried trials and experimenter delivered food trials, even though the highest performances occurred in experimenter delivered food trials in both Sessions 1 and 2. Potentially, the physical proximity of the food and target odor in food buried trials may have enhanced discrimination training, as in these trials the dog recei ved immediate and direct access to food when responding correctly. In the experimenter delivered food trials, the dog had to wait for the experimenter to deliver the food. The potential increase in the delay to the reinforcer in experimenter delivered food trials may negatively impact acquisition, as the delay to reinforcement is known to be an important variable for acquisition (for a review, see Tarpy & Sawabini, 1974). Alternatively, performance may be lowered in the food buried trials, as the odor of t he inaccessible food in the S bin may elicit rooting behaviors that decrease accuracy. Lastly, it is important to note that although dogs were performing significantly above chance in both sessions, performance was marginally higher than chance (60.7%). How this accuracy level may compare to the rate dogs learn a non odor
23 detection task is unclear. Thus, it is unclear if the use of a digging task in pine may facilitate an odor detection task or inhibit performance.
24 Figure 2 1 Layout of Experiment 1 showing e xperim ental bins and a dog r esponding A: Dog making a choice by rooting in one container. B: Left bin is the food accessible bin with the target odor and an open tea ball with treat, right bin is the food inacc essible bin without the target odor and a closed tea ball with treat. C: Dog sniffing and beginning rooting motion. D: The continuation of image C, showing the dog rooting by thrusting the nose into the pine and moving the pine.
25 Figure 2 2. Dog p erfo rmance on Session 1 of Experiment 1. Each dot shows the performance of an individual dog and the height of the bar shows the mean. combined score, and control trials. Error bars show t he 95% CI. The dotted line indicates chance performance. Thus, error bars that do not contain the dotted line show a statistically significant difference from chance on a t test 0 10 20 30 40 50 60 70 80 90 100 Food-buried trials Experimenter-delivered food trials Total Control Percent Correct
26 Figure 2 3 The effects of food deprivation on p erformance in Session 1 of Experiment 1. Each dot indicates the total percent correct in Session 1 for each dog as a function of the number of hours of food deprivation. 0 10 20 30 40 50 60 70 80 90 100 0 5 10 15 20 Total % Correct Hours since last meal
27 Figure 2 4. Dog p erformance on Session 2 of Experiment 1. Each dot shows the performance of an i ndividual dog and the height of the bar shows the mean. combined score, and control trials. Error bars show the 95% CI. The dotted line indicates chance performance. Thus, error bars that do not contain the dotted line show a statistically significant difference from chance on a t test. 0 10 20 30 40 50 60 70 80 90 100 Food-buried trials Experimenter-delivered food trails Total Control Percent Correct
28 CHAPTER 3 EXPERIMENT 2 We designed Experiment 2 to address the questions raised by Experiment 1. In Experiment 2, dogs were given frequent and repeated testing sessions in an odor detection task to assess their rates of learning, the level of performance they can quickly achieve and the stability of the performance of individual dogs. Dogs were given only one trial type (food buried or experimenter delivered food) to assess any differences in the trial types Lastly, dogs were simultaneously trained on a visual discrimination ta sk (white vs. black) using an alternating conditions design to address if the dogs attended to odorant and visual stimuli in the task similarly. Methods Subjects Twenty six pet dogs nave to odor detection training were recruited for participation; however two dogs would not take free food from the experimenter and were not tested. Of the remaining twenty four dogs, twelve dogs were trained using only food buried trials (food buried group) and twelve dogs were trained using only experimenter delivered food trials (experimenter delivered group). All dogs were trained on both the odor detection task from Experiment 1 and a black from white visual discrimination task. Dogs were tested in a familiar environment. Materials Dogs were trained to detect anise scen ted cotton rounds using tan colored (30 cm by 36cm by 15cm ) Sterilite bins (Sterlite Corporation, Townsend, MA) and were trained on the visual discrimination using black and white Sterilite bins (of the same size) filled with pine shavings as specified in Experiment 1. For the odor detection training bins,
29 odors were prepared in the same manner as described in Experiment 1. All general layout procedures not explicitly discussed below were held constant from Experiment 1. Procedures All dogs underwent five testing sessions; testing sessions were spaced between one day and seven days apart. The interval between testing sessions was determined Given the results of Experiment 1, the level of food deprivation was held similar from Session 1 to Session 5 for each dog. Each session consisted of alert training trials, 24 training trials, and 6 control trials. Of the training trials, all dogs underwent a block of 12 odor detection t rials and a block of 12 visual discrimination trials with control trials interspersed every 4 trials. The order of the trial blocks (odor discrimination trials or visual discrimination trials) was counterbalanced across dogs and alternated within dogs from session to session. The target for the visual discrimination (white or black) was counterbalanced across dogs but consistent across sessions within dogs. Food buried group : twelve dogs were trained using only food buried trials from Experiment 1. Immedi ately preceding the block of visual or odor discrimination trials, dogs underwent five alert training trials. Alert training trials were identical to Experiment 1 for the odor detection task, but differed for the visual discrimination task, in that, the b in used for training was the colored target bin (without scented cotton rounds) assigned for that dog (white or black). For the visual discrimination food buried trials, the procedures of the odor detection task were followed except that neither bin was scented with a cotton round and one bin was white whereas the other bin was black. For both the visual and the odor detection trials, food was placed in an open tea ball in the target container and
30 food was placed in a closed (inaccessible) tea ball in th e non target container. If the the dog was rooting in the target container (anise scented for odor, or the target colored If the dog was in the incorrect container, the bins were picked up and the dog was called back. If the dog did not make a choice within 30 seconds, the bins were picked up and re presented. If the dog did not make a choice" was recorded and analyzed as incorrect. All other procedures (e.g. correction trials) were the same as Experiment 1. Experimenter delivered food group : twelve dogs were trained using only experimenter deliv ered food trials from Experiment 1. Immediately preceding all blocks of odor detection training, dogs were given a modified version of the alert training trials. First, a treat was placed on top of the pine shavings and the anise scented pad for two trials Subsequently, dogs were given three trials where the treat and the odor scented round were buried. Once the dog began to root and found the treat, the experimenter trial s, dogs were given three trials of just the scented bin without buried food. Once the a dog failed to root within 30 sec, the experimenter re presented the trial. If t he dog failed to root during an alert training trial, up to two additional trials were given. If dogs failed to root by the additional trials, testing was discontinued. No dogs that rooted when food was buried failed to root by the additional alert trainin g trials. Immediately preceding all blocks of the visual discrimination training, dogs were given the above
31 modified alert training trials except that for the visual discrimination trials, the training bin was the target colored bin (i.e. it was not scente d with anise). For the visual and odor detection trials, no food was buried in the bins. The bins used in subsequent training trials were separate from the alert training trial bins to prevent food odor contamination. During training trials, a naive ob when the observer noted the dog rooting in a container. If the dog was rooting in the dog was rooting in the non target container, the container s were picked up and prepared for the next trial. Control Trials : control trials were the same as control trials in Experiment 1 except that the control trials for the experimenter delivered food group did not have buried food in tea balls. For this group control bins were tan colored and contained only pine shavings. Statistical Analyses An individual criterion of 10 out of 12 correct in a single session was considered significantly above chance (83%, binomial test, p < .04). Group performances were co mpared to chance with a one sample t test. To assess if acquisition was different across the two trial groups (experimenter delivered food group and food buried group) and two task types (odor task and visual task), the cumulative number of correct trials for each session was plotted across the cumulative number of trials for each session, for every dog. For each dog a linear regression line was fit to the data and the rate of acquisition (the slope) was determined. A two way ANOVA was used to compare the a cquisition rate across the four groups and post hoc comparisons were made using a corrected alpha level of 0.013 for multiple comparisons.
32 Results and Discussion Both the food buried group and the experimenter delivered food group performed at chance leve ls during the first 12 trials in both the visual and odor discriminations. A s shown in Figure 3 1 by the end of five sessions, both groups were performing significantly above chance on both discriminations (one sample t test, p largest improve ment was shown in the odor detection task for dogs in the experimenter delivered food group from 53 % to 78% across testing. Dogs in the food buried group showed much less improvement. Performance on the visual task showed modest improvement for both group s, and performance on control trials showed no signs of improvement and remained at chance across all five sessions. For the group showing the greatest improvement (experimenter delivered food on the odor task), a sharp increase in performance was noted f rom Sessions 1 3 with more gradual changes from Session 3 4 and part icularly from Sessions 4 5 ( Figure 3 1). This deceleration in learning once the group reached approximately 78% accuracy is an artifact arising from a sub set of dogs quickly learning the task and performing at high accuracies, while the remaining dogs had not yet acquired the task. Figure 3 2 shows the individual performances across the five sessions for the odor task of the experimenter delivered food group. By the end of Session 3 (36 od or trials), four dogs met the individual criterion of 83%: two of these dogs were p erforming at 100% accuracy ( Figure 3 2). By Session 5, six dogs performed with accuracy levels above 75% accuracy. Figure 3 2 also plots the changes in performance across the five sessions for the eight dogs that achieved the individual criterion for a single session in at least one session. Only five of these dogs, however, maintained accuracy above 83% a cross the
33 last two sessions. Three of these dogs achieved the 92% accuracy across the last two sessions, with one dog achieving 100% accuracy across the last two sessions (Session 4 and 5), one dog achieving 96% accuracy, and one dog achieving 92% accuracy Figure 3 3 shows rates of acquisition by graphing the cumulative number of correct trials by the total number of trials, for each task and trial type combination. The odor task of the experimenter delivered food group shows the highest performance and t he high est rate (slope). In addition, Figure 3 3 shows that dogs start off at the same point, indicating that prior to testing, all dogs performed similarly on both tasks. To assess significant differences between these groups, the slope of each dog for bo th the odor and the visual task was computed from a best fit linear regression line. The individual slopes for each task trial type combination are shown in Figure 3 4. The individual slopes show that three dogs perform at consistently high levels with slo pes greater than .8 (See the open square, filled diamond and filled triangle in Figure 3 2 to see the performance of these dogs across sessions). To compare performance across task and trial type, a two way ANOVA compared the slopes to assess the effect o f the trial type (food buried vs. experimenter delivered) and the task (odor vs. visual). The goodness of fit of the linear regression line for each task trial type combination was good and no systematic variation in the residuals was noted ( r 2 : Odor exper imenter delivered food, .94; Visual experimenter delivered, .95; Odor food buried, .93; Visual food buried, .79). The ANOVA revealed a significant effect of the task, F (1,22) = 20.1, p < .0002, trial type, F (1,22) = 6.34, p < .02, and an interaction, F (1,2 2) = 5.89, p < .02. A paired t test revealed that for the experimenter delivered food group, acquisition was higher for the odor task than the visual task ( t =
34 4.89, df =11, p < .005). An unpaired t test revealed that for the odor task, performance was hi gher in the experimenter delivered food group than the food buried group ( t = 3.54, df =22, p < .002). When considering the effect of trial type on visual task performance, however, no significant difference was found (unpaired t test, t = .93, df = 22, p < .36). In addition, no statistical difference was noted between the task type (odor vs. visual) for the food buried group (paired t test, t = 1.45, df = 11, p < .17). Dogs performed significantly better on the odor task than the visual task, when trained u sing only experimenter delivered food trials. This suggests that the procedure is appropriate for studying odor discrimination learning, as the odor cues provided were readily learned and learned faster than visual cues using the same procedure. The alert training trials and the pine shavings may have prompted sniffing of the odors in the bucket and may have facilitated acquisition of the odor task, however, further testing is needed to elucidate the effects of the pine shavings. Dogs trained with only the experimenter delivered food trials performed significantly better on the odor task than dogs trained with the food buried trials. This runs counter to our hypothesis that physical proximity of the target odor and the accessible reinforcer may reduce the d elay to the reinforcer and enhance discrimination compared to experimenter delivered food trials. It is important to note, however, that food was buried in the S + bin and the S bin, with food only accessible in the S + bin. This was done to insure dogs wer e not simply detecting the smell of food and not the target odor, without requiring additional control trials to test for this possibility. It still remains possible that pairing food only with the target odor may aid training, however, the results from th is experiment suggest it is unlikely. Burying food during the odor task resulted in
35 a 15% decrement in accuracy, indicating that food odor is a powerful stimulus influencing behavior on this task. Food odor may elicit numerous behaviors that may influence overshadow the target odor itself. These results suggest that training without placing the food in close proximity to the odor may be more efficient. It is interesting to note that t here was no difference in performance between the food buried and the experimenter delivered food groups for the visual task. The food odor had no impact on performance for the visual task, indicating that other stimuli were more important in this task.
36 Figure 3 1 Average performance across all five s essions of Experiment 2. Solid lines indicate the dogs in the experimenter delivered food group, dashed lines indicate the food buried group, and the dotted line indicates performance on c ontrol trials. Triangles indicate performance on the odor detection task and circles indicate performance on the visual task. Error bars indicate the standard error. 0 10 20 30 40 50 60 70 80 90 100 Session 1 Session 2 Session 3 Session 4 Session 5 Percent Correct Control Odor (Experimenter-delivered food) Odor (Food buried) Visual (Experimenter-delivered food) Visual (Food buried)
37 Fig ure 3 2. Acquisition of the odor discrimination for e xperimenter d elivered food t rials in Experiment 2. Each symbol shows the performance of an individual dog. Dogs that met the individual criterion of 83% accuracy in at least one session are plotted with a unique symbol to show performance across sessions. Dogs not achieving this cr iterion are graphed with filled circles. Bars show the mean and error bars show the 95% CI. 0 10 20 30 40 50 60 70 80 90 100 Session 1 Session 2 Session 3 Session 4 Session 5 Percent Correct
38 Figure 3 3 Cumulative performance across all t rials. The mean and 95% confidence interval for each task and trial type combination is shown. The odor task of the experimenter delivered food group shows the largest slope and highest performance. Dashed line shows a line with a slope of .5, and would be expected to occur by chance.
39 Figure 3 4 Comparisons of trial types and t asks in Experimen t 2. Dots indi cate dogs in the e xp erimenter delivered food g roup. Squares indicate dogs in the food buried group. Line indicates the mean and error bars show the 95% confidence interval of dogs in each group. indicates a significant differences with a corrected alpha for multiple comparisons. The first row shows the cross subject comparison and the second row shows within subject comparisons. Panels A & B compare the slopes of experimenter delivered food group and food buried group for the odor detection task and the visual task respectively. Panels C & D: compare the slopes of the odor detection task to the visual task for experimenter delivered food group and th e food buried group respectivel y.
40 CHAPTER 4 GENERAL DISCUSSION Results from Experiments 1 and 2 show that dogs can be rapidly trained on an odor detection task. All training, including the training of the alerting response, was carried out during experimentally programmed and recorded trials. This allows comparison ac ross dogs and permits evaluation of variables that may influence the rate at which an odor detection is learnt Experiment 1 demonstrated that dogs can be systematically trained using only discrete trials. Experiment 2 demonstrated the usefulness of the co mponents of this procedure to experimentally evaluate the effects of the different training trials and showed that the rate of acquisition for an odor discrimination was faster than the acquisition of a simple visual discrimination. The first session of E xperiment 1 demonstrated an unexpected negative correlation between hours of food deprivation and rate of acquisition of the odor discrimination; however, this correlation was not replicated during Session 2 of Experiment 1. One limitation to our analysis of food deprivation is that it was an experimental correlate and not an experimentally manipulated variable. Thus, further studies that experimentally manipulate deprivation are required to establish if the observed correlation is a replicable correlation, and if so, whether the correlation represents a causal relationship. Performance on control trials remained at chance levels across all sessions of Experiments 1 and 2. The control trials may have negatively impacted the rate of acquisition, because not presenting the target odor and still reinforcing a response may have interfered with acquisition of the discrimination. It is important, however, to note
41 detection tasks can the odor the dog is detecting (Lit, Schweitzer & Oberbauer 2011). Thus, it is important to control for any potential experimenter cueing. Additionally, it was noted that during control tri als, the highest performing dogs would sniff both buckets and then refrain from responding. On other control trials, the highest performing dogs would sniff both bins and begin to bark at the experimenter or tip over the bins with their paws. Interestingly these dogs would quickly return to rooting appropriately and accurately during the subsequent non control trials. Future studies may utilize control trials to evaluate the effects of the absence of the target odor on behavior during an odor discriminatio n. The procedure utilized for the experimenter delivered food group in Experiment 2 may be useful for identifying individual differences in dogs and selecting for dogs likely to do well in an odor detection training program. Performance varied in this gro up with some dogs quickly learning the odor detection task and performing consistently variable. Dogs that achieve a consistently high level of accuracy may be good dogs to sele ct for further odor detection training. The procedure of Experiment 2 is rapid, requires few materials, and can be administered by individuals with minimal experimental training. The odor task for the experimenter delivered food trials only requires 12 tr ials a session and eight alert training trials. Each trial timed out at 1 minute, indicating that testing took no longer than 20 minutes per session or 100 minutes in total per dog. In addition, if the goal is only to identify dogs that acquire the task mo st rapidly, five sessions may not be necessary. Instead, identify ing only those dogs reaching 83 % accuracy in three or four sessions
42 may be sufficient. The present data suggest, however, that not all dogs reaching the 83% accuracy criterion will maintain t his level of performance consistently over time. Further evaluation is required to determine appropriate selection criteria. The slop e analysis presented in Figure 3 4 may be one promising method for selecting consistently performing dogs. The three dogs w ith slopes great than .8 were consistently high performing dogs across all five sessions. Further evaluation, however, is required to evaluate the slope, or the overall performance on this test, as predictors for how successful a dog will be in a professio nal detector dog training operation. It remains possible that success on this task may not predict success in professional detector dog training operations. One important consideration for this procedure is that dogs were trained to detect the presence of an odor from background pine odor. Dogs were trained on this task as detecting the presence of an odor from background odor may be more similar to real world detection tasks. Alternatively, dogs could be trained to discriminate between two odors, one odor as the S + and a different odor as the S Potentially, utilizing a novel odor as the S instead of only background odor, may enhance the discriminability of the target odor and facilitate learning. This hypothesis, however, requires further testing. Toge ther, the results demonstrate that nave dogs can be trained to detect a novel odor using only discrete trials in a short period of time. The experimenter delivered food procedure in Experiment 2 showed that dogs responded more to odor cues than visual cue s and that consistently high performing dogs on the odor task can be identified within five short testing sessions. This may be a good way to select promising odor
43 detection dogs. The ultimate utility of this procedure in selecting dogs and in studying the variables controlling odor detection will require further evaluatio n.
44 LIST OF REFERENCES Cablk, M. E., Sagebiel, J. C., Heaton, J. S., & Valentin, C. (2008). Olfaction based detection distance: a quantitative analysis of how far away dogs recognize tortoise odor and follow it to source. Sensors 8 (4), 2208 2222. Cornu, J. N., Cancel Tassin, G., Ondet, V., Girardet, C., & Cussenot, O. (2011 ). Olfactory detection of prostate cancer by dogs sniffing urine: a step forward in early diagnosis. European urology 59 (2), 197 201. Dean, E. E. (1972). Training Dogs for Narcotic Detection Retrieved from http://oai.dtic.mil/oai/oai?verb=getRecord&m etad ataPrefix=html&identifier=A 0749302 Fischer Tenhagen, C., Wetterholm, L., Tenhagen, B. A., & Heuwieser, W. (2011). Training dogs on a scent platform for oestrus detection in cows. Applied Animal Behaviour Science 131 (1 2), 63 70. doi:10.1016/j.applanim.20 11.01.006 Furton, K. G., & Myers, L. J. (2001). The scientific foundation and efficacy of the use of canines as chemical detectors for explosives. Talanta 54 (3), 487 500. doi:doi: DOI: 10.1016/S0039 9140(00)00546 4 Goldblatt, A., Gazit, I. & Terkel, J. (2009). Olfaction and explosives detector dogs. In W. Helton, Canine Ergonomics: The science of working dogs (pp. 135 175). Boca Raton, FL: CRC Press. Johnston, J. (1999). Canine detection capabilities: Operational implications of recent R & D findings. Institute for Biological Detection Systems, Auburn University 1 (7). Lit, L., Schweitzer, J. B., & Oberbauer, A. M. (2011). Handler beliefs affect scent detection dog outcomes. Animal cognition 1 8. Maejima, M., Inoue Murayama, M., Tonosaki, K. Matsuura, N., Kato, S., Saito, Y., Weiss, A., et al. (2007). Traits and genotypes may predict the successful training of drug detection dogs. Applied Animal Behaviour Science 107 (3 4), 287 298. National Association for Canine Scent Work (2011). Target O dors. Retrieved from http://www.funnosework.com/targetodors.html Sinn, D. L., Gosling, S. D., & Hilliard, S. (2010). Personality and performance in military working dogs: Reliability and predictive validity of behavioral tests. Applied Animal Behaviour Sc ience 127 (1 2), 51 65. Tarpy, R.M. & Sawabini, F.L. (1974). Reinforcement Delay: a selective review of the last decade. Psychological Bulletin, 81 (12), 984 997. http://ww w.tsa.gov/lawenforcement/programs/puppy_program.shtm
45 Waggoner, L. P. (1998). Effects of extraneous odors on canine detection. Proceedings of SPIE (pp. 355 362). Presented at the Enforcement and Security Technologies, Boston, MA, USA. doi:10.1117/12.335008 Williams, M., & Johnston, J. M. (2002). Training and maintaining the performance of dogs (Canis familiaris) on an increasing number of odor discriminations in a controlled setting. Applied Animal Behaviour Science 78 (1), 55 65. doi:10.1016/S0168 1591(02)0 0081 3 Willis, C. M., Church, S. M., Guest, C. M., Cook, W. A., McCarthy, N., Bransbury, A. J., Church, M. R. T., et al. (2004). Olfactory detection of human bladder cancer by dogs: proof of principle study. B MJ, 329 (7468), 7
46 BIOGRA PHICAL SKETCH Nathaniel Hall was born and raised in St. Petersburg, FL, where he attended Boca Ciega High School and graduated from the Medical Magnet Program. Upon graduation, he was admitted to the University of Florida for undergraduate studies. Thanks to the support of the Florida Opportunity Scholarship, he was able to dedicate his time to his studies in Microbiology and Cell Science while developing an interest in behavioral research. After conducting an undergraduate thesis on the behavior of bats, Nathaniel continued with behavioral research. He graduated in the spring of 2010 with a Bachelor of Science majoring in microbiology and cell science. Nathaniel subsequently entered a doctoral program in Psychology. Nathaniel plans to continue his re search and hopes t o obtain his PhD in Psychology