Chapter 4 Hysteresis and adaptation

Status: manuscript and first stage registered report in preparation; presented as conference posters at ECVP 2019 and SIPS 2019 (Van Geert & Wagemans, 2019a, 2019b)

Chapter 3 supported the idea that prägnant steps on a dimension can influence percept formation and consequently task performance, as evidenced by differences in categorization performance and response time as well as the category boundary and within-category boundary effects in discrimination performance and perceived similarity. Besides longer-term prägnant steps as reference points, also the immediate context in which a structure is presented may play a role in how this structure is perceived. In this chapter, we explore potential effects of immediate perceptual history and stimulus history on percept formation.

Earlier research has found both attractive and repulsive effects of temporal context (Snyder et al., 2015). With hysteresis, we refer to the attractive effect of a previous percept on the current percept. With adaptation, we refer to the repulsive effect of a previous stimulus on the current percept (see Figure 4.1). Whereas hysteresis could help to stabilize a system and avoid focus on changes in distracting, unnecessary details (cf. simplification, leveling), adaptation could help to notice and emphasize relevant and characteristic changes (cf. complication, sharpening).

Figure 4.1: Illustration of attractive and repulsive context effects. Left: attraction effect (hysteresis). When the stimulus is perceived as a car at time 1 (T1), the probability that another stimulus at time 2 (T2) will be perceived as a car is higher than when the stimulus at T1 was interpreted as a tortoise. Right: repulsion effect (adaptation). When the stimulus at T1 is a very clear example of a car, the probability that another stimulus at T2 will be perceived as a car is lower than when the stimulus at T1 was a more ambiguous example of a car.

You may notice the similarity of the concept pair hysteresis and adaptation to the concept pair of robustness and sensitivity to change in Chapter 3. Although also hysteresis and adaptation seem contradictory at first sight, they are not complete opposites and work complementary as well. The five ways to reconcile robustness and sensitivity to change mentioned in Chapter 3 can also lead to insights in the relation between hysteresis and adaptation.

4.1 Hysteresis and adaptation in categorization, discrimination, and similarity judgment tasks

We reanalyzed the data presented in Chapter 3 to determine the presence or absence of hysteresis and adaptation effects in the categorization, discrimination, and similarity judgment task. Because the amount of data per subject was limited, only an analysis across participants could be conducted.

A clear hysteresis effect was present in case of the non-identifiable morph series, for all tasks (see Figures D.1a, D.2a, and D.3a). Indications of an adaptation effect were present for the non-identifiable morph series in the categorization task only (Figure D.1b). In the discrimination and similarity judgment task, no conclusive evidence for an adaptation effect was found (Figures D.2b and D.3b).

We could conclude that if no clear prägnant steps are available (i.e., in the non-identifiable morph series), temporal context information may become more important in disambiguating the incoming visual stimulation. The previous percept may then function as a sort of reference point to which the incoming input is drawn.

The hysteresis and adaptation effects we describe here for the categorization task are average results across participants. Does every individual show these hysteresis and adaptation effects in perception, however, and does every individual show these effects to the same extent?

4.2 Individual differences in hysteresis and adaptation effects on the perception of non-identifiable morph figures

To further investigate the presence of individual differences in hysteresis and adaptation effects, we conducted a new study with more data per participant, and focusing on the non-identifiable morph series and the categorization task only. The results from the reanalysis were replicated: across individuals, a hysteresis and an adaptation effect were present. This adaptation effect interacted with the hysteresis effect and the current morph level, however.

In the case of an ambiguous current stimulus, the hysteresis effect was larger when the previous stimulus did not carry much evidence for either of the two categories. A typical adaptation effect was present when the previous percept was in line with the evidence in the previous stimulus. When the previous percept was incongruent with the evidence in the previous stimulus (which might have been a response error in some cases), an opposite adaptation effect was present: the stronger the evidence in the previous stimulus, i.e., the more incongruent the previous stimulus and response/percept were, the more often the current response/percept was in line with the evidence in the previous stimulus.

When the current stimulus was a rather clear example of a category (A/B) and the previous percept (B/A) was incongruent with the evidence in the current stimulus, the more evidence there was for the reported percept in the previous stimulus, the more often the current stimulus was perceived as the opposite category (i.e., adaptation effect).

When the current stimulus was a rather clear example of a category (A/B) and the previous percept (A/B) was congruent with the evidence in the current stimulus, the more evidence there was for the reported percept in the previous stimulus, the more often the current stimulus was perceived as the same category (i.e., opposite adaptation effect).

The adaptation effect diminished in the second block of the experiment, when the categorization performance also increased slightly. This finding is in line with our hypothesis based on the reanalysis described above, that context information would play a bigger role in situations where there are no clear prägnant steps available yet. It also gives more direct evidence for a decrease in the use of context information when categorization is stronger and hence more prägnant Gestalts are formed.

All effects were similar in both morph series used, but categorization was a bit weaker in one of the morph series, and this also seemed to result in a slightly bigger hysteresis effect (see D.5).

As our main interest was investigating individual differences, we conducted a Bayesian multilevel logistic regression analysis on the data. Although the results indicate that everyone shows an hysteresis effect, not everyone showed an adaptation effect (Figure 4.2a). Some participants even showed an opposite adaptation effect. Individual estimates of hysteresis and adaptation effects seemed to be highly correlated (Figure 4.2b): those individuals who show a stronger (i.e., more negative) adaptation effect also show a stronger hysteresis effect.

Figure 4.2: (a) Individual slope estimates for the hysteresis and adaptation effect based on a Bayesian multilevel logistic regression model. (b) Correlation of individual slope estimates presented in (a). Note. Black dots and grey lines indicate the median and 95% highest density continuous interval per individual. The black and the blue line indicate the null value and the average median across participants respectively. In the interactive version of this document, hover over the data points to see the exact values for each individual participant.

Although the morph level had the intended effect for almost all participants (some had an effect close to zero), the model also indicated large differences between individuals in strength of categorization, i.e., how strongly an individual’s percept depends on the current morph level presented. The model indicated a strong relation between the effect of the current morph level and individual estimates of the adaptation effect: the stronger an individual is influenced by the current morph level, the stronger also the influence of the previous morph level on the current percept. The model also indicated a slight positive relation between the effect of the current morph level and individual estimates of the hysteresis effect: the stronger the effect of the current morph level, the stronger the effect of the previous percept on the current percept.

In addition, we were interested in the relation of individual’s hysteresis and adaptation strength with other, more high-level differences between individuals: (a) an individual’s need for closure, tolerance for ambiguity, and intolerance of uncertainty; (b) an individual’s sensory sensitivity for visual stimuli; (c) the extent to which an individual shows schizotypic traits; (d) an individual’s vividness of visual imagery; and (e) the extent to which an individual demonstrates autistics traits. All questionnaire data except for the autism-related questionnaire were collected in a separate research session. Number of participants for which the correlations could be conducted thus varied between 130 and 209. None of the more high-level constructs had a strong relation with the size of the individual’s behavioral categorization, hysteresis, or adaptation effects (see Figure D.9).

4.3 Individual differences in hysteresis and adaptation effects on the perception of multistable dot lattices

Gepshtein & Kubovy (2005) presented an elegant paradigm that allows to disentangle attractive and repulsive context effects on perception. They used multistable dot lattices (see Figures 4.3 and 4.4) as context and test stimuli, and investigated the influence of (a) the perceived organization of the context stimulus (i.e., which organization was reported) and (b) the stimulus support for a certain organization in the context stimulus (dependent on the stimulus’ aspect ratio) on the perception of a second, test stimulus.

Figure 4.3: Explanation regarding the aspect ratio of a multistable rectangular dot lattice. In rectangular dot lattices, four different orientations can be perceived, of which two are more prevalent (as the dots are closer together along these orientations). The relative dominance of the a orientation relative to the b orientation is expressed in the aspect ratio of the dot lattice (AR = |a| / |b|).

Figure 4.4: Aspect ratio of a multistable rectangular dot lattice.

We assume this paradigm is ideal to study individual differences in hysteresis and adaptation in a more controlled way than we did in the previous experiment because (a) average hysteresis and adaptation effects have already been established using this paradigm (cf. Gepshtein & Kubovy, 2005; Schwiedrzik et al., 2014), (b) the paradigm uses more ‘low-level’ stimuli where more high-level influences may be less prominent, and (c) the difference between different stimulus levels can more easily be quantified in this paradigm.

I am currently preparing a First Stage Registered Report that will investigate whether the average hysteresis and adaptation effects, as well as their non-interaction, presented by Gepshtein & Kubovy (2005) and Schwiedrzik et al. (2014), can be replicated; whether everyone shows both effects; whether interindividual differences exist in the size of the effects; and if individual differences in hysteresis and adaptation exist, whether these correlate with another behavioral parameter, more specifically the individual’s absolute orientation bias. For the latter, we expect that individuals who show a stronger absolute orientation bias (i.e., individuals for whom some orientations are strong prägnant steps) will be less influenced by previously shown stimuli and previously experienced percepts.

Different from the previously described study, this study cannot investigate the relation between the effect of the current stimulus and hysteresis and adaptation effects, as the second stimulus in this paradigm is always ambiguous.

Although the data collection for this project can only start after submission and in-principle-acceptance of the First Stage Registered Report, I can present preliminary results based on the data I received from the study by Schwiedrzik et al. (2014). In this dataset, everyone showed a hysteresis effect, and almost everyone showed an adaptation effect (see Figure D.14). The effects of hysteresis and adaptation were positively correlated (Figure D.15): in general, the more hysteresis an individual showed, the more adaptation that person showed as well.

4.4 Future research ideas

4.4.1 Review on hysteresis and adaptation

During the preparation of the studies above, I noticed how fragmented the literature concerning hysteresis and adaptation is. Both hysteresis (e.g., attractive context effect, serial dependency, history effect, recency bias, perceptual stability, assimilation, contraction) and adaptation (e.g., contrastive context effect, repulsion, adaptation, perceptual warping, differentiation) are referred to in a multitude of ways.

As indicated in the introduction of this chapter, I posit that the Prägnanz literature can inspire ideas of how the seemingly contrasting effects of hysteresis and adaptation can be reconciled. One possible future research idea is to prepare a thorough literature review on the scattered literature concerning hysteresis and adaptation and their interaction. This literature review could in its turn inspire new empirical tests (e.g., variants of the empirical paradigms mentioned above) to disentangle both effects.

4.4.2 Generalizability of findings

In experimental psychology, paradigms are often either created very ad hoc, or they are used because they have shown to be succesful in the past. The concepts we use are much broader than each of the paradigms used, however. Are the findings made with one such paradigm also generalizable to other task contexts, or are we fooling ourselves by keeping fishing in te same pond? To investigate the generalizability of the empirical findings in 4.3, we propose to not only assess individual hysteresis and adaptation effects in the dot lattice paradigm specified above, but to also measure individual differences in these effects in more standard hysteresis (i.e., ordered presentation) and adaptation (i.e., more extreme first lattice shown for a longer time) paradigms. Also temporal and spatial tilt aftereffect tasks may be included in the data collection, to investigate the generalizability of the results to other stimuli and to the spatial rather than temporal domain.

4.4.3 Interaction with longer-term context effects

A future study may investigate the interaction between short-term temporal context effects such as hysteresis and adaptation with longer-term temporal context effects. For example, a context-setting phase could preceed the dot lattice paradigm specified in 4.3. In this context-setting phase, participants could be more often presented with a specific absolute orientation. We could then study the effect of this longer-term difference in absolute orientations presented in the task context on the shorter-term hysteresis and adaptation effects: will the overall probability of perceiving the primed orientation diminish the hysteresis and/or the adaptation effect?