When we perceive a stimulus, our organization of that stimulus does not only depend on the stimulus itself, but also on the temporal context (in which the stimulus is presented) AND on the individual that is processing the stimulus in that context.

Earlier research found two different temporal context effects.

First, they found a tendency to organize the visual input in a similar way as preceding context stimuli. We will call this attractive effect of the previous percept "hysteresis".

Second, they found a tendency that repels/pushes the current percept away from the organization that was most dominant in the previous stimulus. We will call this repulsive effect of the previous stimulus "adaptation".

Until now these context effects have mainly been studied on a group level. In our study, we investigated whether true and consistent individual differences exist in the extent to which their perception is influenced by the immediate context, and if so, whether every individual does show both context effects in the expected directions.

Like earlier research, we used multistable dot lattices as stimuli.

In rectangular dot lattices, two dominant orientations can be perceived. By varying the distance between the dots in one compared to the other direction, one orientation will be easier to perceive than the other (this is a manipulation of the aspect ratio).

Here you see three examples: chances are higher to perceive the right-tilted a orientation in the left stimulus, and the left-tilted b orientation in the right stimulus. In the middle stimulus, probabilities for perceiving a and b are equal.

In hexagonal dot lattices, three orientations are equally dominant. We could thus say that hexagonal lattices are often more ambiguous than rectangular dot lattices.

How did we then study the effect of context using these lattices?

In each trial, we first presented participants with a rectangular dot lattice. In this first lattice, aspect ratio was manipulated across trials. Participants reported which orientation they perceived by choosing from four possible orientations. After response, they were presented with a hexagonal dot lattice, and they again reported which orientation they perceived.

Participants came for two sessions one to two weeks apart, and in every session they conducted 90 trials per aspect ratio. The absolute orientation of the lattice was randomly varied across trials (90° in steps of 1°).

We were interested in investigating the influence of the percept of the first lattice on the percept of the second lattice, as well as the influence of the aspect ratio in the first lattice on how the second lattice was perceived.

On the left you can get an idea of how the task looked like for participants [WAIT!].

What do we get out of conducting this paradigm?

From previous research we know that both context effects are found in this paradigm when averaging across participants.

The graph on the right shows the aspect ratio of the first lattice on the x axis and logit probability of perceiving orientation a in the second lattice, separately for when the first lattice was perceived as a or as b.

The vertical distance between the line of black and the line of grey dots indicates the attractive effect of the previous percept: if orientation a was perceived in the first lattice, chances were higher to also perceive orientation a in the second lattice.

The positive slope of both lines indicates the adaptation effect: the less evidence there was for orientation a in the first stimulus, the more chance of perceiving orientation a in the second stimulus.

[In other words, having perceived orientation a in the previous stimulus, increases the chances of perceiving the same orientation again in the next stimulus. On the other hand, been presented with a stimulus containing a lot of evidence for the a orientation decreases the chances of perceiving orientation a in the next (more ambiguous/multistable) stimulus.]

We performed this study as a Registered Report, and besides many other research questions, asked whether individuals differ in how they use immediate context information in their perception.

Our focus was on whether there are individual differences in the size and/or the direction of these context effects, and whether these individual differences were stable across time.

We fitted several Bayesian logistic hierarchical models of perceived L2 orientation, and used Bayes Factors to compare different models in light of our research questions.

With the Bayes Factor, we compare the likelihood of the observed data under different models. In this way, we answer the question under which of the two models the observed data are more likely to occur.

Let's now have a look at some results of the study.

As expected, we replicated the results of the previous studies on the group level: In the perception of the second lattice, there was an attractive effect of the previous percept (difference between the two lines in the graph) as well as a repulsive effect of the aspect ratio of the first stimulus (visible based on the positive slope of the lines).

Furthermore, concerning the first lattice, we see a direct effect of aspect ratio on the perceived orientation, so a direct effect of grouping by proximity.

But what about the results at the individual level?

We found true individual differences in the size of both hysteresis and adaptation effects.

Furthermore, almost everyone showed effects in the expected direction, although our Bayes Factors indicate that not everyone shows true effects in the expected direction.

The number of participants with an mean estimated true non-positive hysteresis or adaptation effects was very low.

The strong positive correlation between estimated individual effects for the first and second session. Both for hysteresis and for adaptation effects, there is a strong positive correlation between the estimated effects for the first and second session, which indicates temporal consistency of the individual differences over a period of 7 to 14 days.

One additional result I want to mention has to do with a control task participants conducted in the first session. In a separate task, participants were presented with random dot lattices as first lattice in which no actual orientation could be perceived. This excludes the possibility for a perceptual influence of having seen an orientation in the first lattice, but keeps the decisional influence of having responded with a certain orientation after being presented with the first lattice.

By comparing the estimated hysteresis effects of the participants in both tasks, we can get an idea of whether hysteresis is a purely decisional effect, a purely perceptual effect, or a combination of both.

On average a hysteresis effect was present in both tasks, but the hysteresis effect was significantly smaller in the control task than in the experimental task. Therefore, we conclude that hysteresis is partly perceptual and partly decisional. The results nuance earlier perspectives stating serial dependence to be either a fully percept-related or a fully decision-related effect.

Furthermore, from the left graph we can suggest that individuals seem to differ in the extent to which their hysteresis effect is percept- or decision-related: a considerable number of participants did not show evidence for a hysteresis effect in the control task, hinting at a more perceptual basis for their hysteresis effects.

One aspect of the study that we did only explore after registering the study was whether we could also find consistent individual differences in the direct effect of aspect ratio on the perception of the first lattice.

We found strong evidence for consistent individual differences in the true effect of proximity on individual's perception of multistable dot lattices, and slight evidence that every individual has a proximity effect in the expected direction.

Similar to the temporal context effects, also the individual proximity effects showed a strong positive correlation across sessions. The proximity effect also seemed to slightly increase in the second session.

What can we conclude from this study?

In this study, we replicated the average attractive effect of the previous percept on the current percept, and the repulsive effect of previously presented stimulus evidence on the current percept.

Large individual differences in the size of these effects exist, and these individual differences are consistent across one to two weeks time.

Although almost everyone shows both effects in the expected direction, not every single individual does.

Besides these main analyses, we investigated much more, and if you want to know more I encourage you to read the paper published in i-Perception.

[For example, based on comparing the hysteresis effect in the experimental and a control task, we can conclude that hysteresis is partially percept- and partially decision- or response-related. Moreover, also the direct proximity effect on perception of the first lattice varied consistently across individuals, there is slight evidence that everyone shows the proximity effect in the expected direction, and the individual differences in proximity showed stable across sessions.]

As a key takeaway, this study provides insight in how individuals differ in how they combine previous input and experience with current input in their perception, and more generally, this teaches us that different individuals can perceive identical stimuli differently, even when presented in a similar context.

I want to thank you all for your attention and point you to the paper and all additional resources available online.

To get to all of those links, you can go to the url shown on the bottom of the slide and click all the links on the slides. p If there are any questions, I am more than happy to answer them :-)

We fitted several Bayesian logistic hierarchical model of perceived L2 orientation, and used Bayes Factors to compare different models in light of our research questions.

What does this Bayes factor comparison entail? More specifically, we compare the likelihood of the observed data under different models, to answer the question under which of the two models the observed data are more likely to occur.

For example, when investigating whether the size of temporal context effects differs between individuals, we compare the likelihood of the observed data under the unconstrained model (that is the model placing no order or equality constraints on individuals' effects) and under the common effect model (that is the model assuming that everyone has the same true effect size). If we then find a Bayes Factor larger than one, this indicates that the observed data are more likely under the unconstrained model than under the common effect model.

Similarly, when assessing whether everyone shows the effects in the expected direction, we compared the likelihood of the observed data under the positive effects model (that is a model assuming everyone has a true effect in the expected direction) and under the unconstrained model. A Bayes Factor above 1 indicates that the data are more likely under the positive effects model than under the unconstrained model.

[We decided to take a Bayes Factor of 6 or 1/6 as convincing evidence for the existence of individual differences in size or direction of these effects.]

When correlating individual estimates for both temporal context effects, we find a strong positive correlation. Note that as the estimated correlation value comes from a hierarchical model including both estimated effects, potential attenuation of the correlation as a result of noise is already taken into account. This strong positive correlation suggests that there may be at least a common factor affecting the processes underlying both temporal context effects.

As we wanted to investigate the relation of the size of longer-term biases to more often perceive certain orientations and the size of the shorter-term temporal context effects, we added an absolute orientation bias task to the study in which participants respond which orientation they perceive for a series of multistable hexagonal dot lattices.

When correlating the strength of individuals' absolute orientation bias with their estimated hysteresis and adaptation effects, we did not find a clear linear relation, especially not a negative one as we originally expected. Although the high Bayes Factors for the non-negative model indicate that the observed data are more likely under a non-negative linear relationship than under a model assuming a negative linear relationship, this does not mean that the winning model necessarily provides a good fit with the data. Although we tested whether a quadratic model could better predict the data pattern, the Bayes Factor was in favor of the linear model compared to the quadratic one.

The results thus provide slight evidence for a positive relationship between long- and short-term biases. Also the linear model does not seem to provide a good fit for the data.