- Email: [email protected]
Integration of multiple memories in sensory preconditioning
Behavioural Processes 108 (2014) 94–97
Contents lists available at ScienceDirect
Behavioural Processes journal homepage: www.elsevier.com/locate/behavproc
Short report
Integration of multiple memories in sensory preconditioning Paul Craddock a,∗ , Charlotte Renaux a , Franc¸oise Lefèvre a , James Byron Nelson b , Mikael Molet a a b
University of Lille, North of France, France University of the Basque Country, Spain
a r t i c l e
i n f o
Article history: Received 18 February 2014 Received in revised form 1 September 2014 Accepted 26 September 2014 Available online 5 October 2014 Keywords: Reaction time Sensory preconditioning Human conditioning
a b s t r a c t The present study demonstrates that humans’ response to a single stimulus (S1) is determined by what follows S1’s associates. The experiment used a sensory preconditioning (SPC) design where S1 was associated with both S2 and S3 on separate trials before establishing relationships between these latter stimuli with an outcome or its absence in a second phase. When S2 and S3 were associated with the same consequence, either an outcome or its absence, strong consequence-based responding to S1 was observed in a reaction time test. Participants responded quickly to indicate that the outcome was, or was not, predicted by S1. When S2 predicted the outcome and S3 did not, SPC was weaker although participants were not slower to respond to S1. Implications on the understanding of the mechanisms that allow for the response to S1 to emerge are discussed. © 2014 Elsevier B.V. All rights reserved.
1. Introduction Sensory preconditioning (SPC; e.g., Brogden, 1939) consists of pairing two neutral stimuli (S1–S2) in phase 1 of an experiment and then pairing S2 with an unconditioned stimulus (US; S2-US) in phase 2. In this paradigm, a conditioned response (CR) appropriate to the US is observed to S1, even though S1 itself has never been directly paired with the US. Such a response is indicative of the integration (Matzel et al., 1988; Savastano & Miller, 1998; Arcediano & Miller, 2002) of the Phase 1 and Phase 2 memories. It is as if memories can be superimposed upon each other allowing for the response to S1 to emerge. A possible explanation is that S1 evokes an associative chain of memories at test (S1 → S2 → US) leading to the response (see Molet et al., 2012). Alternatively, S2 could evoke S1 during Phase 2, allowing a direct association to form between the representation of S1 and the US (e.g., see Hall, 1996; Holland, 2008). In a conventional SPC procedure, S1 only has a single associate, S2. In a natural setting, stimuli are likely to have multiple associates, each of which can be associated with yet other stimuli. For example, S1 might be associated with both S2 and S3. Subsequently, S2 might be predictive of the US, while S3 is not. In such a
∗ Corresponding author at: Université de Lille, Nord de France, Domaine universitaire du “Pont de Bois”, rue du Barreau, BP 60149, 59653 Villeneuve d’Ascq Cedex, France. Tel.: +33 328481551. E-mail address: [email protected] (P. Craddock). http://dx.doi.org/10.1016/j.beproc.2014.09.034 0376-6357/© 2014 Elsevier B.V. All rights reserved.
situation, weaker SPC is expected compared to a condition in which S2 and S3 are both predictive of the US, either because S1 retrieved two conflicting associations regarding the US, or because the representation of S1 was paired with the US only in the presence of S2. We assume that if the weaker SPC is explained solely by the activation of two conflicting associations, the time to respond that S1 predicts the US should be increased in comparison to conditions in which S2 and S3 consistently predict the US or the absence of the US. The goal of the present study is two-fold. First, it attempted to examine the role of reaction time in the understanding of the mechanisms underlying SPC. Second, it sought to further validate a recently developed procedure which combines both response time and type of response when participants make predictions concerning S1’s outcome (Craddock et al., 2012). 2. Experiment The experiment was designed to assess how memories controlled by multiple associates of a single stimulus combine using a SPC design. In the first phase, S1 was paired with both S2 and S3 (simultaneous pairings; e.g., Recorla, 1980), on different trials in all conditions. Then, in a Consistent Outcome condition S2 and S3 were each paired with the same outcome stimulus (sequential pairings) in phase 2. Here, we would expect that S1 should elicit a response appropriate to expecting the outcome. In a No Outcome condition, S2 and S3 were presented in phase 2, but followed by nothing. A response to S1 on test appropriate to expecting no outcome is expected in this condition. The most interesting data
P. Craddock et al. / Behavioural Processes 108 (2014) 94–97
95
Table 1 Design. Condition
S2–S1 phase (simultaneous pairings)
S2/S3-Outcome and S2/S3-No outcome phase (sequential pairings)
Test
Consistent outcome
S1–S2 (Color1 + Letter1) S1–S3 (Color1 + Letter2)
S2-Outcome (Letter1 − Outcome) S3-Outcome (Letter2 − Outcome)
Color1? Letter1? Letter2?
Inconsistent outcome
S1–S2 (Color2 + Letter3) S1-S3 (Color2 + Letter4)
S2-Outcome (Letter3 − Outcome) S3-No outcome (Letter4 − No-outcome)
Color2? Letter3? Letter4?
No outcome
S1–S2 (Color3 + Letter5) S1–S3 (Color3 + Letter6)
S2-No outcome (Letter5 − No-outcome) S3-No outcome (Letter6 − No-outcome)
Color3? Letter5? Letter6?
will come from a third condition, which is the Inconsistent Outcome condition. During phase 2, S2 was followed by the outcome, while S3 was not. On test, S1 should yield lower associative strength with the outcome compared to the Consistent Outcome condition, and higher associative strength than in the No Outcome condition because of S2 and S3’s conflicting outcomes. We used a recently developed transformation rule to evaluate associative strength (Craddock et al., 2012). A strong association between events is hypothesized to result in a quick decision that the second event will be present. Weaker associations are assumed to be reflected in more uncertainty (i.e., longer response times and/or inconsistent predictions). The absence of association between a cue and an outcome is assumed to be reflected in a quick prediction that the second event will be absent. The experiment was conducted within subjects with the different conditions outlined above created with different stimuli. The design is shown in Table 1. Colored squares (Color1–Color3) served as S1 stimuli. Letters (Letter1–Letter6) served as S2 and S3 stimuli. A picture of two rabbits served as the outcome. In the Consistent Outcome condition, Color1 was associated with both Letter1 and Letter2, which each predicted the outcome in phase 2. In the No Outcome condition, Color3 was associated with Letter5 and Letter6, which each predicted the absence of the outcome in phase 2. In the Inconsistent Outcome condition Color2 was associated with Letter3 and Letter4 and one of these stimuli was associated with the outcome and the other with the absence of the outcome in phase 2. On test the S1–S3 stimuli were presented and participants were required to respond “yes” or “no” by pressing keys on a keyboard to indicate whether the stimulus presented predicted the outcome or not. 2.1. Method 2.1.1. Participants Sixteen French university students (8 males and 8 females; 18–29 years of age) volunteered. 2.1.2. Materials The stimulus presentations were programmed using Python 2.7. Light gray randomly selected Latin letters in Helvetica 350 bold font were presented on a dark gray background of a computer screen. Three 8 cm × 8 cm blue, red, or green squares were used as the three color-stimuli mentioned in Table 1. The 2-s outcome stimulus was the one used by Craddock et al. (2012) and consisted of a vertical yellow stripe (19 cm × 7.5 cm) in the middle of the screen on which was drawn two rabbits. 2.1.3. Procedure Before starting the experiment, each of the six letters was randomly assigned to one of the three colors. Phase 1 began with the computer instructing participants that they were to learn colorletter associations. Participants then received eight blocks of 2-s
simultaneous pairings of S1 (colors) with each of S2 and S3 (Letters) in each condition with the order of the six stimulus presentations randomized within each block and between participants. The intertrial intervals were 1.5 s. Each letter appeared either on the left (50% of the participants) or the right half of the computer screen. The colored squares were presented on the opposite side. For each participant, the letters always appeared on the same side of the screen. At the start of phase 2, participants were instructed to learn which letters were followed by an image of two rabbits (Outcome). Afterwards, the letters appeared for 2-s in the same position on the screen as during Phase 1. There were six blocks of six randomized presentations of the letters either immediately followed by the outcome stimulus (image of two rabbits) or by no outcome (a blank screen) depending on the condition to which the letters were assigned. The inter-trial intervals were 1.5-s. The only time participants were to respond was during the test. Participants were informed that they would have to indicate whether each stimulus predicted the rabbits. Participants were not told that their reaction time (RT) was being recorded but they were invited to respond as soon as they knew their answer. The instructions indicated that the first trial was for training: a symbol (‘&’) rather than a letter was tested. The test stimulus remained on the screen until the participant had responded then ‘next trial’ was displayed for 1.5-s. Participants indicated their choice by pressing the left or right arrow keys (counterbalanced as “yes” or “no” between subjects). Each of the six letters and the three color-stimuli was tested in a random order, in the same screen position they had during learning. The type of response (i.e., ‘yes’ or ‘no’) and the RT to respond was recorded. Each of the nine RTs was divided by the largest RT of the nine (RTmax). Any ‘yes’ response RT was then transformed into [−ln(RT/RTmax)], and any ‘no’ response RT into [+ln(RT/RTmax)]. Hence, transformed data near zero (i.e., RT/RTmax near 1) denote slow, hesitating, responses, while positive or negative transformed data indicate that the participant answered ‘yes’ or ‘no’ without hesitation, respectively (see Craddock et al., 2012). Because a slight increase or decrease in a short RT is presumably more significant than an equivalent increase or decrease in a long RT (e.g., a difference of 2-s between two RTs is relatively more important when the RTs are short than when they are long), differences between short RTs were made more important than differences between long RTs, through the logarithmic transformation. 2.2. Results The results were analyzed using nonparametric statistics (i.e., Wilcoxon test) due to some distributions not being normal. The effect sizes were calculated as Cohen’s g where g = 0.5(# (xi1 < xi2 ) − # (xi1 > xi2 ))/n, (i.e., .one-half of Cliff’s d. Cliff’s d corresponds to the proportion of times when the difference between two repeated measures is strictly negative minus the
P. Craddock et al. / Behavioural Processes 108 (2014) 94–97
RT transformed data
96
3.0 2.5 2.0 1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 -2.0 -2.5 -3.0 -3.5 -4.0
Consistent Outcome (Color1)
Inconsistent Outcome (Color2)
No Outcome (Color3)
Condition Fig. 1. Boxplot of the transformed reaction-time data. The black and gray boxes represent the interquartile ranges. The change of color within the boxes corresponds to the median. Vertical lines encompass the range. The transformed data correspond to the absolute value of the natural logarithms of each ratio RT/RTmax to which the signs ‘+’ or ‘−’ are allocated depending on whether the response is ‘yes’ or ‘no’, respectively.
proportion of times when this difference is strictly positive; Cliff, 1993; Grissom & Kim, 2005). Effect sizes measured using Cohen’s g are categorized as small (.05 – .15), medium (.15–.25) and large (>25) with a maximum of 0.5 (Cohen, 1988). 2.2.1. Type of response and reaction time Most participants (i.e., 11 out of 16) decided that Color2 was predictive of the outcome, suggesting that learning about the outcome may have been slightly more powerful than learning about no outcome. The Mac Nemar test shows that the proportion of ‘yes’ responses was greater for Color2 than for Color3 (2 (1) = 7.36; p < .01), showing that SPC took place despite the conflicting predictions of S1’s associates. No difference in the proportion of ‘yes’ responses was evidenced between Color2 and Color1 (i.e., 11 out of 16 compared to 13 out of 16). Interestingly, there was no difference in response times either between Color1 and Color2 (z = 0.66, p = .50), or between Color2 and Color3 (z = 1.02, p = .30). A detailed observation of data for Color2 showed that ‘yes’ responses were slow (median = 8.64 s compared to 4.27 s for the ‘yes’ responses of Color1), while ‘no’ responses were fast (median = 1.81 s compared to 11.06 s for Color1’s ‘no’ responses). Taken together, these results suggest that SPC is not uniquely explained by the activation of S1’s associates and their respective associations with the outcome, because Color2, despite having associates predicting opposite outcomes did not generate slower responses. 2.2.2. Transformed data In the present paragraph, reaction times and type of response are integrated in a single measure according to the rule presented above. Letter1, Letter2, and Letter3 were followed by the outcome and, consequently, produced similar rapid positive responding (medians = 1.42, 1.69, and 1.65, respectively; 2 (2) = 0.88; Friedman test; p > .60), while Letter4, Letter5, and Letter6 were paired with no outcome and generated similar rapid negative responding (medians = −1.56, −1.44, and −1.46, respectively; 2 (2) = 2.38; Friedman test; p > .30). In the Inconsistent Outcome condition the reinforced S2 (i.e., Letter3) was more predictive of the outcome stimulus than S3 (i.e., Letter4) which was paired with no outcome (z = 3.5, [Wilcoxon test], p = .0005; g = .5), showing first-order conditioning.
Interestingly, every participant made correct predictions for Letter3 and Letter4, suggesting that if the test of S1 (i.e., Color2) activated the representation of its associates (i.e., Letter3 and Letter4), and their respective associative strengths with the outcome, conflicting predictions should have lengthened response times when Color2 was tested. Fig. 1 depicts the response to S1 (i.e., the color cues) between the conditions on test. SPC was demonstrated by comparing the predictive potentials of S1 between the Consistent Outcome and No Outcome conditions. S1 (i.e., Color1) was associated with both S2 and S3 stimuli that were both predictive of the outcome in the Consistent Outcome condition while S1 (i.e., Color3) was associated with S2 and S3 that were never followed by the outcome in the No Outcome condition. A Wilcoxon test revealed that S1 was more strongly associated with the presence of the outcome stimulus in the former condition over the latter condition (z = 3.5, p = .0005; g = .5). Of most interest, responding to S1 in the Inconsistent Outcome condition was less than in the Consistent Outcome condition, (i.e., scores to Color2 were smaller than scores to Color1; z = 2.52, p = .012; g = .31) and marginally greater than the No Outcome condition (i.e., scores for Color2 were marginally greater than those for Color3; z = 1.9, p = .057; g = .25). The transformation rule seems to adequately integrate response time and type of response to provide intermediate values for SPC when S1’s associates make opposite predictions. The weaker SPC in that condition could only be evidenced by integrating RTs and type of responses in one single measure. 3. Discussion The experiment assessed the efficiency of the transformation rule introduced by Craddock et al. (2012) in a modified SPC procedure. In all conditions, S1 was associated with both S2 and S3. In the Consistent Outcome condition, S2 and S3 were both associated with an outcome. In the No Outcome condition, S2 and S3 were associated with no outcome. In the Inconsistent Outcome condition, S2 but not S3 was associated with an outcome. The results showed that stimuli in a SPC procedure can have multiple associates, and each of those stimuli can be associated with yet other stimuli. Moreover, although the experiment cannot lead to a clear interpretation of the processes underlying SPC, it does provide information on how the different memories combine.
P. Craddock et al. / Behavioural Processes 108 (2014) 94–97
When tested with S1, appropriate responding was greater in the Consistent Outcome and No Outcome conditions. The Inconsistent Outcome condition produced intermediate responding to S1. One interpretation of SPC is that the representation of S1 directly associates with the outcome during the S2-outcome pairings of Phase 2. Another interpretation is that S1, at test, activates the representation of S2 and its association with the outcome. The latter explanation, applied to the present design, implies that S1 activated both the representation of S2 and of S3 and their opposite predictions concerning the outcome, in the Inconsistent Outcome condition. Because the representations involved conflicting predictions, it was assumed that the response time would be longer in that condition compared to the two other conditions, particularly because each participant accurately predicted the outcome following S2 and the absence of outcome following S3. However, the global results did not show such increased response times in the Inconsistent Outcome condition. Nevertheless, ‘no’ responses for Color2 were fast (median = 1.81 s) compared to the three ‘no’ responses to Color1 that were slow (median = 11.06 s). This may suggest that SPC was less readily obtained in the Inconsistent Outcome condition. Moreover, ‘yes’ responses in that condition were slow (median = 8.64 s). Taken together these results suggest that SPC was slowly learned in the Inconsistent Outcome condition compared to the Consistent Outcome condition (i.e., quick ‘no’ responses in the former condition), presumably because the representation of S1 was less often directly paired with the outcome in the former condition compared to the latter condition (i.e., the representation of S1 was paired with the outcome only during the S2-outcome trials). It can also be assumed that, once SPC was established, conflicting memories interacted (i.e., RTs to ‘yes’ responses to Color2 were long), presumably because S1 activated the representation of its associates in order to determine the type of response. It is only the integration of these two measures (i.e., reaction time and type of responses) with the transformation rule that can
97
display the expected linear evolution across the three conditions studied here (Fig. 1). However, the independent study of each of the two dependent variables gives a glimpse on the processes underlying SPC. Acknowledgments Participation of James Byron Nelson was supported in part by grant PSI2011-24231 from the Spanish Ministry of Science and Innovation and grant IT-694-13 from the Basque Ministry of Science. References Arcediano, F., Miller, R.R., 2002. Some constraints for models of timing: a temporal coding hypothesis perspective. Learn. Motiv. 33, 105–123. Brogden, W.J., 1939. Sensory pre-conditioning. J. Exp. Psychol. 25, 323–332. Cliff, N., 1993. Dominance statistics: ordinal analyses to answer ordinal questions. Psychol. Bull. 114, 494–509. Cohen, J., 1988. Statistical Power Analysis for the Behavioral Sciences, 2nd ed. Academic Press, New York. Craddock, P., Molet, M., Miller, R.R., 2012. Reaction time as a measure of human associative learning. Behav. Process. 90, 189–197. Grissom, R.J., Kim, J.J., 2005. Effect Sizes for Research. A Broad Practical Approach. Erlbaum, Mahwah, NJ. Hall, G., 1996. Learning about associatively activated representations: implications for acquired equivalence and perceptual learning. Anim. Learn. Behav. 24, 233–255. Holland, P., 2008. Cognitive versus stimulus-response theories of learning. Learn. Behav. 36, 227–241. Matzel, L.D., Held, F.P., Miller, R.R., 1988. Information and expression of simultaneous and backward associations: implications for contiguity theory. Learn. Motiv. 19, 317–344. Molet, M., Miguez, G., Cham, H.X., Miller, R.R., 2012. When does integration of independently acquired temporal relationships take place? J. Exp. Psychol. Anim. Behav. Process. 38, 369–380. Recorla, R.A., 1980. Simultaneous and successive associations in sensory preconditioning. J. Exp. Psychol. Anim. Behav. Process. 6, 207–216. Savastano, H.I., Miller, R.R., 1998. Time as content in Pavlovian conditioning. Behav. Process. 44, 147–162.
Contents lists available at ScienceDirect
Behavioural Processes journal homepage: www.elsevier.com/locate/behavproc
Short report
Integration of multiple memories in sensory preconditioning Paul Craddock a,∗ , Charlotte Renaux a , Franc¸oise Lefèvre a , James Byron Nelson b , Mikael Molet a a b
University of Lille, North of France, France University of the Basque Country, Spain
a r t i c l e
i n f o
Article history: Received 18 February 2014 Received in revised form 1 September 2014 Accepted 26 September 2014 Available online 5 October 2014 Keywords: Reaction time Sensory preconditioning Human conditioning
a b s t r a c t The present study demonstrates that humans’ response to a single stimulus (S1) is determined by what follows S1’s associates. The experiment used a sensory preconditioning (SPC) design where S1 was associated with both S2 and S3 on separate trials before establishing relationships between these latter stimuli with an outcome or its absence in a second phase. When S2 and S3 were associated with the same consequence, either an outcome or its absence, strong consequence-based responding to S1 was observed in a reaction time test. Participants responded quickly to indicate that the outcome was, or was not, predicted by S1. When S2 predicted the outcome and S3 did not, SPC was weaker although participants were not slower to respond to S1. Implications on the understanding of the mechanisms that allow for the response to S1 to emerge are discussed. © 2014 Elsevier B.V. All rights reserved.
1. Introduction Sensory preconditioning (SPC; e.g., Brogden, 1939) consists of pairing two neutral stimuli (S1–S2) in phase 1 of an experiment and then pairing S2 with an unconditioned stimulus (US; S2-US) in phase 2. In this paradigm, a conditioned response (CR) appropriate to the US is observed to S1, even though S1 itself has never been directly paired with the US. Such a response is indicative of the integration (Matzel et al., 1988; Savastano & Miller, 1998; Arcediano & Miller, 2002) of the Phase 1 and Phase 2 memories. It is as if memories can be superimposed upon each other allowing for the response to S1 to emerge. A possible explanation is that S1 evokes an associative chain of memories at test (S1 → S2 → US) leading to the response (see Molet et al., 2012). Alternatively, S2 could evoke S1 during Phase 2, allowing a direct association to form between the representation of S1 and the US (e.g., see Hall, 1996; Holland, 2008). In a conventional SPC procedure, S1 only has a single associate, S2. In a natural setting, stimuli are likely to have multiple associates, each of which can be associated with yet other stimuli. For example, S1 might be associated with both S2 and S3. Subsequently, S2 might be predictive of the US, while S3 is not. In such a
∗ Corresponding author at: Université de Lille, Nord de France, Domaine universitaire du “Pont de Bois”, rue du Barreau, BP 60149, 59653 Villeneuve d’Ascq Cedex, France. Tel.: +33 328481551. E-mail address: [email protected] (P. Craddock). http://dx.doi.org/10.1016/j.beproc.2014.09.034 0376-6357/© 2014 Elsevier B.V. All rights reserved.
situation, weaker SPC is expected compared to a condition in which S2 and S3 are both predictive of the US, either because S1 retrieved two conflicting associations regarding the US, or because the representation of S1 was paired with the US only in the presence of S2. We assume that if the weaker SPC is explained solely by the activation of two conflicting associations, the time to respond that S1 predicts the US should be increased in comparison to conditions in which S2 and S3 consistently predict the US or the absence of the US. The goal of the present study is two-fold. First, it attempted to examine the role of reaction time in the understanding of the mechanisms underlying SPC. Second, it sought to further validate a recently developed procedure which combines both response time and type of response when participants make predictions concerning S1’s outcome (Craddock et al., 2012). 2. Experiment The experiment was designed to assess how memories controlled by multiple associates of a single stimulus combine using a SPC design. In the first phase, S1 was paired with both S2 and S3 (simultaneous pairings; e.g., Recorla, 1980), on different trials in all conditions. Then, in a Consistent Outcome condition S2 and S3 were each paired with the same outcome stimulus (sequential pairings) in phase 2. Here, we would expect that S1 should elicit a response appropriate to expecting the outcome. In a No Outcome condition, S2 and S3 were presented in phase 2, but followed by nothing. A response to S1 on test appropriate to expecting no outcome is expected in this condition. The most interesting data
P. Craddock et al. / Behavioural Processes 108 (2014) 94–97
95
Table 1 Design. Condition
S2–S1 phase (simultaneous pairings)
S2/S3-Outcome and S2/S3-No outcome phase (sequential pairings)
Test
Consistent outcome
S1–S2 (Color1 + Letter1) S1–S3 (Color1 + Letter2)
S2-Outcome (Letter1 − Outcome) S3-Outcome (Letter2 − Outcome)
Color1? Letter1? Letter2?
Inconsistent outcome
S1–S2 (Color2 + Letter3) S1-S3 (Color2 + Letter4)
S2-Outcome (Letter3 − Outcome) S3-No outcome (Letter4 − No-outcome)
Color2? Letter3? Letter4?
No outcome
S1–S2 (Color3 + Letter5) S1–S3 (Color3 + Letter6)
S2-No outcome (Letter5 − No-outcome) S3-No outcome (Letter6 − No-outcome)
Color3? Letter5? Letter6?
will come from a third condition, which is the Inconsistent Outcome condition. During phase 2, S2 was followed by the outcome, while S3 was not. On test, S1 should yield lower associative strength with the outcome compared to the Consistent Outcome condition, and higher associative strength than in the No Outcome condition because of S2 and S3’s conflicting outcomes. We used a recently developed transformation rule to evaluate associative strength (Craddock et al., 2012). A strong association between events is hypothesized to result in a quick decision that the second event will be present. Weaker associations are assumed to be reflected in more uncertainty (i.e., longer response times and/or inconsistent predictions). The absence of association between a cue and an outcome is assumed to be reflected in a quick prediction that the second event will be absent. The experiment was conducted within subjects with the different conditions outlined above created with different stimuli. The design is shown in Table 1. Colored squares (Color1–Color3) served as S1 stimuli. Letters (Letter1–Letter6) served as S2 and S3 stimuli. A picture of two rabbits served as the outcome. In the Consistent Outcome condition, Color1 was associated with both Letter1 and Letter2, which each predicted the outcome in phase 2. In the No Outcome condition, Color3 was associated with Letter5 and Letter6, which each predicted the absence of the outcome in phase 2. In the Inconsistent Outcome condition Color2 was associated with Letter3 and Letter4 and one of these stimuli was associated with the outcome and the other with the absence of the outcome in phase 2. On test the S1–S3 stimuli were presented and participants were required to respond “yes” or “no” by pressing keys on a keyboard to indicate whether the stimulus presented predicted the outcome or not. 2.1. Method 2.1.1. Participants Sixteen French university students (8 males and 8 females; 18–29 years of age) volunteered. 2.1.2. Materials The stimulus presentations were programmed using Python 2.7. Light gray randomly selected Latin letters in Helvetica 350 bold font were presented on a dark gray background of a computer screen. Three 8 cm × 8 cm blue, red, or green squares were used as the three color-stimuli mentioned in Table 1. The 2-s outcome stimulus was the one used by Craddock et al. (2012) and consisted of a vertical yellow stripe (19 cm × 7.5 cm) in the middle of the screen on which was drawn two rabbits. 2.1.3. Procedure Before starting the experiment, each of the six letters was randomly assigned to one of the three colors. Phase 1 began with the computer instructing participants that they were to learn colorletter associations. Participants then received eight blocks of 2-s
simultaneous pairings of S1 (colors) with each of S2 and S3 (Letters) in each condition with the order of the six stimulus presentations randomized within each block and between participants. The intertrial intervals were 1.5 s. Each letter appeared either on the left (50% of the participants) or the right half of the computer screen. The colored squares were presented on the opposite side. For each participant, the letters always appeared on the same side of the screen. At the start of phase 2, participants were instructed to learn which letters were followed by an image of two rabbits (Outcome). Afterwards, the letters appeared for 2-s in the same position on the screen as during Phase 1. There were six blocks of six randomized presentations of the letters either immediately followed by the outcome stimulus (image of two rabbits) or by no outcome (a blank screen) depending on the condition to which the letters were assigned. The inter-trial intervals were 1.5-s. The only time participants were to respond was during the test. Participants were informed that they would have to indicate whether each stimulus predicted the rabbits. Participants were not told that their reaction time (RT) was being recorded but they were invited to respond as soon as they knew their answer. The instructions indicated that the first trial was for training: a symbol (‘&’) rather than a letter was tested. The test stimulus remained on the screen until the participant had responded then ‘next trial’ was displayed for 1.5-s. Participants indicated their choice by pressing the left or right arrow keys (counterbalanced as “yes” or “no” between subjects). Each of the six letters and the three color-stimuli was tested in a random order, in the same screen position they had during learning. The type of response (i.e., ‘yes’ or ‘no’) and the RT to respond was recorded. Each of the nine RTs was divided by the largest RT of the nine (RTmax). Any ‘yes’ response RT was then transformed into [−ln(RT/RTmax)], and any ‘no’ response RT into [+ln(RT/RTmax)]. Hence, transformed data near zero (i.e., RT/RTmax near 1) denote slow, hesitating, responses, while positive or negative transformed data indicate that the participant answered ‘yes’ or ‘no’ without hesitation, respectively (see Craddock et al., 2012). Because a slight increase or decrease in a short RT is presumably more significant than an equivalent increase or decrease in a long RT (e.g., a difference of 2-s between two RTs is relatively more important when the RTs are short than when they are long), differences between short RTs were made more important than differences between long RTs, through the logarithmic transformation. 2.2. Results The results were analyzed using nonparametric statistics (i.e., Wilcoxon test) due to some distributions not being normal. The effect sizes were calculated as Cohen’s g where g = 0.5(# (xi1 < xi2 ) − # (xi1 > xi2 ))/n, (i.e., .one-half of Cliff’s d. Cliff’s d corresponds to the proportion of times when the difference between two repeated measures is strictly negative minus the
P. Craddock et al. / Behavioural Processes 108 (2014) 94–97
RT transformed data
96
3.0 2.5 2.0 1.5 1.0 0.5 0.0 -0.5 -1.0 -1.5 -2.0 -2.5 -3.0 -3.5 -4.0
Consistent Outcome (Color1)
Inconsistent Outcome (Color2)
No Outcome (Color3)
Condition Fig. 1. Boxplot of the transformed reaction-time data. The black and gray boxes represent the interquartile ranges. The change of color within the boxes corresponds to the median. Vertical lines encompass the range. The transformed data correspond to the absolute value of the natural logarithms of each ratio RT/RTmax to which the signs ‘+’ or ‘−’ are allocated depending on whether the response is ‘yes’ or ‘no’, respectively.
proportion of times when this difference is strictly positive; Cliff, 1993; Grissom & Kim, 2005). Effect sizes measured using Cohen’s g are categorized as small (.05 – .15), medium (.15–.25) and large (>25) with a maximum of 0.5 (Cohen, 1988). 2.2.1. Type of response and reaction time Most participants (i.e., 11 out of 16) decided that Color2 was predictive of the outcome, suggesting that learning about the outcome may have been slightly more powerful than learning about no outcome. The Mac Nemar test shows that the proportion of ‘yes’ responses was greater for Color2 than for Color3 (2 (1) = 7.36; p < .01), showing that SPC took place despite the conflicting predictions of S1’s associates. No difference in the proportion of ‘yes’ responses was evidenced between Color2 and Color1 (i.e., 11 out of 16 compared to 13 out of 16). Interestingly, there was no difference in response times either between Color1 and Color2 (z = 0.66, p = .50), or between Color2 and Color3 (z = 1.02, p = .30). A detailed observation of data for Color2 showed that ‘yes’ responses were slow (median = 8.64 s compared to 4.27 s for the ‘yes’ responses of Color1), while ‘no’ responses were fast (median = 1.81 s compared to 11.06 s for Color1’s ‘no’ responses). Taken together, these results suggest that SPC is not uniquely explained by the activation of S1’s associates and their respective associations with the outcome, because Color2, despite having associates predicting opposite outcomes did not generate slower responses. 2.2.2. Transformed data In the present paragraph, reaction times and type of response are integrated in a single measure according to the rule presented above. Letter1, Letter2, and Letter3 were followed by the outcome and, consequently, produced similar rapid positive responding (medians = 1.42, 1.69, and 1.65, respectively; 2 (2) = 0.88; Friedman test; p > .60), while Letter4, Letter5, and Letter6 were paired with no outcome and generated similar rapid negative responding (medians = −1.56, −1.44, and −1.46, respectively; 2 (2) = 2.38; Friedman test; p > .30). In the Inconsistent Outcome condition the reinforced S2 (i.e., Letter3) was more predictive of the outcome stimulus than S3 (i.e., Letter4) which was paired with no outcome (z = 3.5, [Wilcoxon test], p = .0005; g = .5), showing first-order conditioning.
Interestingly, every participant made correct predictions for Letter3 and Letter4, suggesting that if the test of S1 (i.e., Color2) activated the representation of its associates (i.e., Letter3 and Letter4), and their respective associative strengths with the outcome, conflicting predictions should have lengthened response times when Color2 was tested. Fig. 1 depicts the response to S1 (i.e., the color cues) between the conditions on test. SPC was demonstrated by comparing the predictive potentials of S1 between the Consistent Outcome and No Outcome conditions. S1 (i.e., Color1) was associated with both S2 and S3 stimuli that were both predictive of the outcome in the Consistent Outcome condition while S1 (i.e., Color3) was associated with S2 and S3 that were never followed by the outcome in the No Outcome condition. A Wilcoxon test revealed that S1 was more strongly associated with the presence of the outcome stimulus in the former condition over the latter condition (z = 3.5, p = .0005; g = .5). Of most interest, responding to S1 in the Inconsistent Outcome condition was less than in the Consistent Outcome condition, (i.e., scores to Color2 were smaller than scores to Color1; z = 2.52, p = .012; g = .31) and marginally greater than the No Outcome condition (i.e., scores for Color2 were marginally greater than those for Color3; z = 1.9, p = .057; g = .25). The transformation rule seems to adequately integrate response time and type of response to provide intermediate values for SPC when S1’s associates make opposite predictions. The weaker SPC in that condition could only be evidenced by integrating RTs and type of responses in one single measure. 3. Discussion The experiment assessed the efficiency of the transformation rule introduced by Craddock et al. (2012) in a modified SPC procedure. In all conditions, S1 was associated with both S2 and S3. In the Consistent Outcome condition, S2 and S3 were both associated with an outcome. In the No Outcome condition, S2 and S3 were associated with no outcome. In the Inconsistent Outcome condition, S2 but not S3 was associated with an outcome. The results showed that stimuli in a SPC procedure can have multiple associates, and each of those stimuli can be associated with yet other stimuli. Moreover, although the experiment cannot lead to a clear interpretation of the processes underlying SPC, it does provide information on how the different memories combine.
P. Craddock et al. / Behavioural Processes 108 (2014) 94–97
When tested with S1, appropriate responding was greater in the Consistent Outcome and No Outcome conditions. The Inconsistent Outcome condition produced intermediate responding to S1. One interpretation of SPC is that the representation of S1 directly associates with the outcome during the S2-outcome pairings of Phase 2. Another interpretation is that S1, at test, activates the representation of S2 and its association with the outcome. The latter explanation, applied to the present design, implies that S1 activated both the representation of S2 and of S3 and their opposite predictions concerning the outcome, in the Inconsistent Outcome condition. Because the representations involved conflicting predictions, it was assumed that the response time would be longer in that condition compared to the two other conditions, particularly because each participant accurately predicted the outcome following S2 and the absence of outcome following S3. However, the global results did not show such increased response times in the Inconsistent Outcome condition. Nevertheless, ‘no’ responses for Color2 were fast (median = 1.81 s) compared to the three ‘no’ responses to Color1 that were slow (median = 11.06 s). This may suggest that SPC was less readily obtained in the Inconsistent Outcome condition. Moreover, ‘yes’ responses in that condition were slow (median = 8.64 s). Taken together these results suggest that SPC was slowly learned in the Inconsistent Outcome condition compared to the Consistent Outcome condition (i.e., quick ‘no’ responses in the former condition), presumably because the representation of S1 was less often directly paired with the outcome in the former condition compared to the latter condition (i.e., the representation of S1 was paired with the outcome only during the S2-outcome trials). It can also be assumed that, once SPC was established, conflicting memories interacted (i.e., RTs to ‘yes’ responses to Color2 were long), presumably because S1 activated the representation of its associates in order to determine the type of response. It is only the integration of these two measures (i.e., reaction time and type of responses) with the transformation rule that can
97
display the expected linear evolution across the three conditions studied here (Fig. 1). However, the independent study of each of the two dependent variables gives a glimpse on the processes underlying SPC. Acknowledgments Participation of James Byron Nelson was supported in part by grant PSI2011-24231 from the Spanish Ministry of Science and Innovation and grant IT-694-13 from the Basque Ministry of Science. References Arcediano, F., Miller, R.R., 2002. Some constraints for models of timing: a temporal coding hypothesis perspective. Learn. Motiv. 33, 105–123. Brogden, W.J., 1939. Sensory pre-conditioning. J. Exp. Psychol. 25, 323–332. Cliff, N., 1993. Dominance statistics: ordinal analyses to answer ordinal questions. Psychol. Bull. 114, 494–509. Cohen, J., 1988. Statistical Power Analysis for the Behavioral Sciences, 2nd ed. Academic Press, New York. Craddock, P., Molet, M., Miller, R.R., 2012. Reaction time as a measure of human associative learning. Behav. Process. 90, 189–197. Grissom, R.J., Kim, J.J., 2005. Effect Sizes for Research. A Broad Practical Approach. Erlbaum, Mahwah, NJ. Hall, G., 1996. Learning about associatively activated representations: implications for acquired equivalence and perceptual learning. Anim. Learn. Behav. 24, 233–255. Holland, P., 2008. Cognitive versus stimulus-response theories of learning. Learn. Behav. 36, 227–241. Matzel, L.D., Held, F.P., Miller, R.R., 1988. Information and expression of simultaneous and backward associations: implications for contiguity theory. Learn. Motiv. 19, 317–344. Molet, M., Miguez, G., Cham, H.X., Miller, R.R., 2012. When does integration of independently acquired temporal relationships take place? J. Exp. Psychol. Anim. Behav. Process. 38, 369–380. Recorla, R.A., 1980. Simultaneous and successive associations in sensory preconditioning. J. Exp. Psychol. Anim. Behav. Process. 6, 207–216. Savastano, H.I., Miller, R.R., 1998. Time as content in Pavlovian conditioning. Behav. Process. 44, 147–162.