Timing matters: The temporal representation of experience in subjective mood reports
This article has been Reviewed by the following groups
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
- Evaluated articles (eLife)
Abstract
Humans refer to their own mood state regularly in day-to-day as well as in clinical interactions. Theoretical accounts suggest that when reporting on our mood we integrate over the history of our experiences; yet, the temporal structure of this integration remains unexamined. Here we use a computational approach to quantitatively answer this question and show that early events exert a stronger influence on the reported mood compared to recent events. We show that a Primacy model accounts better for mood reports compared to a range of alternative temporal representations, and replicate this result across random, consistent or dynamic structures of reward environments, age groups and both healthy and depressed participants. Moreover, we find evidence for neural encoding of the Primacy, but not the Recency, model in frontal brain regions related to mood regulation. These findings hold implications for the timing of events in experimental or clinical settings and suggest new directions for individualized mood interventions.
Significance
How we rate our own mood at any given moment is shaped by our experiences; but are the most recent experiences the most influential, as assumed by current theories? Using several sources of experimental data and mathematical modeling, we show that earlier experiences within a context are more influential than recent events, and replicate this finding across task environments, age groups, and in healthy and depressed participants. Additionally, we present neural evidence supporting this primacy model. Our findings show that delineating a temporal structure is crucial in modeling mood and this has key implications for its measurement and definition in both clinical and everyday settings.
Article activity feed
-
###Reviewer #3:
This is an interesting study in which the authors compare Primacy and Recency weighting models' ability to predict momentary mood assessments during a well-established gambling task. They do so across a range of conditions:
i) random/structured/structured-adaptive reward environments
ii) different age groups
iii) in healthy versus depressed participants They also perform the same task in fMRI. They find that the Primacy model wins in most cases, and relates more strongly to brain activations in fMRI.
The paper is very clearly written and easy to read and understand. The conclusions are striking, given the greater dominance of recency-based models in the literature (e.g. Kahneman's peak-end heuristic). I do however have some major concerns with some aspects of the modelling and task design: I'm not sure if they are …
###Reviewer #3:
This is an interesting study in which the authors compare Primacy and Recency weighting models' ability to predict momentary mood assessments during a well-established gambling task. They do so across a range of conditions:
i) random/structured/structured-adaptive reward environments
ii) different age groups
iii) in healthy versus depressed participants They also perform the same task in fMRI. They find that the Primacy model wins in most cases, and relates more strongly to brain activations in fMRI.
The paper is very clearly written and easy to read and understand. The conclusions are striking, given the greater dominance of recency-based models in the literature (e.g. Kahneman's peak-end heuristic). I do however have some major concerns with some aspects of the modelling and task design: I'm not sure if they are addressable or not. In summary, they are:
i) the comparison of Primacy and Recency models doesn't seem fair to me, as the models also differ according to whether the E term is based on previous expectations or previous outcomes. How can the authors conclude that primacy/recency is the key feature of the winning model?
ii) The structured and structured-adaptive versions of the task seem to me to have potential biases against the Recency model due to confounding effects: these other effects must be excluded for the conclusions to be robust.
The following describes these and other concerns in more detail:
Methods:
The modelling seems to me to be problematic as a contrast between primacy and recency because the Primacy and Recency models differ in more than one respect: not just weighting of previous events (presented as the "critical difference between the two models" on p6), but also whether those events are expectations (in the Recency model) or outcomes (in the Primacy model). If the authors want to conclusively establish that Primacy is a better model than Recency then surely more models ought to be compared, at very least using a 2x2 design with primacy/recency of expectations/outcomes? This is also an issue for the fMRI analysis: it is hard to conclude much about the models from the fact that the Primacy model E beta (but not the Recency model E beta) correlates with a BOLD cluster when the Recency model E term is based on previous expectations, not previous outcomes. Likewise with the direct comparison of the models' voxel-wise correlation images.
There also seems to be an error in Figure 1's Equation (1): presumably this just refers to the Primacy model's E term and not the Recency model's E term? Both should be shown for clarity. Also Equation (6) does not look like Equation (1) - is Equation (6) incorrect? In which case what is the R term supposed to look like in Equation (6) - is it also subject to primacy weighting or not? Also in the Discussion, the authors say the Primacy model maintained the overall exponential discounting of the E term. I might misunderstand but this seems a bit misleading because the discounting is by γ^(t-j) in one model but γ^k in the other?
The authors also comment that the Primacy model performed better "when we did not distinguish between gambling and non-gambling trials, which was another divergence from the standard Recency model". But as I understand it, the standard Recency model was originally designed such that the certain option C was NOT the average of the two gambles, so C was required in the model (at least in the 2014 PNAS paper). Here, C is the average of the gambles, so presumably it would be identical to E in the Recency model, and therefore be extraneous in the Recency model as well as the Primacy model - did the authors do model comparison to see if it could be eliminated from the Recency model? If so, this is not another difference between the models after all. Apologies if I have misunderstood something...
I might be misunderstanding the fitting approach here but it sounds like the leave-out sample validation is done to optimise the hyperparameters, not the parameters? In which case there is no complexity penalty to reduce overfitting in the plain MSE measure? I appreciate this is less of an issue if models have the same number of parameters...
Results:
The authors state that the Primacy model does best in the Random condition but this is not what is stated in Table S1, where its MSE is higher, not lower (0.006 vs 0.0008)?
A major issue with the task structures as they stand is that the structured and structured-adaptive tasks seem to have some potential problems when it comes to assessing their impact on mood ratings:
i) the valence of the blocks was not randomised, meaning that the results could be confounded by valence. E.g., what if negative RPE effects are longer-lasting than positive RPE effects? This seems plausible given the downward trend in mood in the random environment despite an average RPE of zero. This could also explain the pattern of mood in the other two tasks, rather than primacy?
ii) issues of scale: if there is a non-linear relationship between cumulative RPE and mood, such that greater and greater RPEs are required to lift/decrease mood by the same amounts, then this will resemble a primacy effect? This is unlikely to be an issue in the random task but may well be a problem in the structured and certainly in the structured-adaptive tasks?
iii) issues of individual differences in responsiveness to RPE: in the structured-adaptive task, some subjects' mood ratings may be very sensitive to RPE, and others very insensitive. One might expect that given the control algorithm has a target mood, the former group would reach this target fairly soon and then have trials without RPE, and the latter group would not reach the target despite ever increasing RPEs. In both cases the Primacy model would presumably win, due to sensitivity to outcomes in the first half or insensitivity to bigger outcomes in the second half respectively? Can these possibilities be excluded using model comparison methods?
These issues are a concern because the plain MSE is not an ideal model comparison method, and the Streaming Prediction MSE is equivocal between the Primacy and Recency models in the Random environment - the only environment which seems unbiased towards the models (given the adolescent sample was also Structured-Adaptive).
-
###Reviewer #2:
In this paper the authors report data from a series of online and one neuroimaging study in which participants played a simple game in which they had to select between a sure outcome and a gamble. Participants reported their current mood throughout the game and the authors compared the performance of a number of models of how the mood ratings were generated. They focus on two models, a standard model which assumes that participants' expectations assume a 50:50 gamble and an adapted model that uses average experienced outcomes as the expected value. They frame these models in terms of recency vs. past weighting and suggest that the results provide evidence in favour of a higher weight of earlier events on reported mood.
The question of how humans combine experienced events into reported mood is topical. This paper takes …
###Reviewer #2:
In this paper the authors report data from a series of online and one neuroimaging study in which participants played a simple game in which they had to select between a sure outcome and a gamble. Participants reported their current mood throughout the game and the authors compared the performance of a number of models of how the mood ratings were generated. They focus on two models, a standard model which assumes that participants' expectations assume a 50:50 gamble and an adapted model that uses average experienced outcomes as the expected value. They frame these models in terms of recency vs. past weighting and suggest that the results provide evidence in favour of a higher weight of earlier events on reported mood.
The question of how humans combine experienced events into reported mood is topical. This paper takes an interesting approach to this issue.
I struggled a bit to understand the logic of some of the arguments in the paper, in part because important experimental and methodological detail is missing. I list my points below. The overriding question is, I think, how certain we can be that the results reported by the authors reflect a true primacy effect, as opposed to some other process (e.g. just learning an expected value) that appears in this case to be a primacy effect.
I didn't really understand where the weights from the primacy graph in Figure 1B came from. The recency weights make sense-there is a discount factor in the model that is less than 1, so there is an exponential discount of more distant past events. However, for the primacy model the expectation is calculated as the mean (apparently arithmetic mean) of previous outcomes (which suggests a flat weight across previous trials) and the discount factor remains-so how does this generate the decreasing pattern of weights? It would be really useful if the authors could spell this out.
The models seem to differ in terms of whether they learn about the expected value of the gamble outcomes or whether they assume a 50:50 gamble (the recency model assumes this, the primacy model generates an average of all experienced outcomes). Might the benefit of the primacy model when explaining human behaviour simply be that people use experienced outcomes to generate their expectations rather than taking stated outcome probabilities as absolutes? In other words, it is not so much that people place more weight on earlier events, but that they learn.
Linked to the above, the structured and adaptive environments seem to have something to learn (blocks with positive vs. negative RPEs), so it is perhaps not surprising that humans show evidence of learning here and a model with some learning outperforms one with none. The description of these environments isn't really sufficient at present-please explain how RPEs were manipulated (was it changing the probability of win/loss outcomes, if so, how? Or was it changing the magnitude of the options? For the adaptive design was the change deterministic? So was the outcome, and thus RPE, always positive if mood was low, or was this probabilistic and if so with what probability?). Also, did the recency model still estimate its expectations here as 50:50, even when (if) this was not the case? If so, can the authors justify this?
What were participants told about the gambles (i.e. were they told they were 50:50, including in structured/adaptive environments)?
Please report the estimated parameter values of the models (and tell us where the common parameters differed between models). This would help in understanding how they are behaving.
In addition to changing the expectation term of the recency model, the primacy model also drops the term of for the sure outcomes (because this improves the performance of the primacy model). Does this account for the relative advantage of the primacy over the recency model? i.e. if the sure outcome term is dropped from the recency model, does the primacy model still perform better?
-
###Reviewer #1:
Keren and co. presents a very interesting study whose goal is to determine what are the determinants of subjective mood rating. They correctly identify as the "baseline" model the model proposed by Rutledge et al. where a big determinant of mood seems to be the reward prediction error (Recency model) and they contrast it with a Primacy model, where first events (not late events) play a more important role.
They validate the model across different behavioural datasets, involving (supposedly) healthy subjects, teenagers and depressive patients. They also have a fMRI experiment and found that the weights of the Primacy model (and not the weights of the Recency model) correlate across subjects with prefrontal activity.
Overall I think this paper addresses an important question and presents an impressive amount of data. …
###Reviewer #1:
Keren and co. presents a very interesting study whose goal is to determine what are the determinants of subjective mood rating. They correctly identify as the "baseline" model the model proposed by Rutledge et al. where a big determinant of mood seems to be the reward prediction error (Recency model) and they contrast it with a Primacy model, where first events (not late events) play a more important role.
They validate the model across different behavioural datasets, involving (supposedly) healthy subjects, teenagers and depressive patients. They also have a fMRI experiment and found that the weights of the Primacy model (and not the weights of the Recency model) correlate across subjects with prefrontal activity.
Overall I think this paper addresses an important question and presents an impressive amount of data. However, I do believe that there are some important checks to be made both concerning the computational and the fMRI analyses.
Concerning model comparison, I would like the authors to show us whether or not their model selection criteria allows us to correctly recover the true generative model in simulated datasets. Are we sure that the model selection criteria are unbiased toward the two models?
Equally important: can the authors provide at the group level a qualitative signature of mood data that falsify the Recency model (see Palminteri, Wyart and Koechlin. 2017). They do so in Figure S2 for one subject, but it would be important to show the same (or similar) result at the group level. This should be easier in the structured or in the structured-adaptive conditions.
Concerning neuroimaging, if I am not missing something, the results they present in the main texte is the results of a second level ANCOVA, where the individual weights of the Primacy model are shown to correlate with activity in the prefrontal cortex. Similar analyses using the weights of Recency model do not produce significant results at the chosen threshold. This analysis is problematic for two reasons. First, absence of evidence does not imply evidence of absence. Second, to really validate the model the authors should show that the trial-by-trial correlates of expectations and prediction errors are consistent with the Primacy and not the recency model. Can the authors show that the Primacy regressors explain better trial-by-trial neural activity compared to the competing model? They could do so formally by estimating the model using the Baysian toolbox usually used to compare DCM models.
Also concernant neuroimaging, I would be important to verify that the authors replicate Rutledge et al's results and Vinckier et al's results (vmPFC, insula, striatum...). This will tell us if the studies are really comparable and would be informative regardless of the result.
-
##Preprint Review
This preprint was reviewed using eLife’s Preprint Review service, which provides public peer reviews of manuscripts posted on bioRxiv for the benefit of the authors, readers, potential readers, and others interested in our assessment of the work. This review applies only to version 2 of the manuscript.
###Summary:
This is a very interesting study whose goal is to determine what drives subjective mood over time during a reward-based decision making task. The authors report data from a series of online studies and one performed with fMRI. Participants played a well-established gambling task during which they had to select between a sure outcome and a 50:50 gamble, reporting momentary mood assessments throughout the game. The authors compared the performance of a number of models of how the mood ratings were generated.
The…
##Preprint Review
This preprint was reviewed using eLife’s Preprint Review service, which provides public peer reviews of manuscripts posted on bioRxiv for the benefit of the authors, readers, potential readers, and others interested in our assessment of the work. This review applies only to version 2 of the manuscript.
###Summary:
This is a very interesting study whose goal is to determine what drives subjective mood over time during a reward-based decision making task. The authors report data from a series of online studies and one performed with fMRI. Participants played a well-established gambling task during which they had to select between a sure outcome and a 50:50 gamble, reporting momentary mood assessments throughout the game. The authors compared the performance of a number of models of how the mood ratings were generated.
The authors identify as their "baseline" model that proposed by Rutledge and colleagues, in which an important determinant of mood seems to be the reward prediction error: the authors call this Recency model. They contrast it with a Primacy model, where earlier events (in this case, average experienced outcomes) play a more important role. They validate the model across different behavioural conditions, involving healthy subjects, teenagers and depressive patients. The conclusion is that the data are more consistent with their Primacy model, in other words a higher weight of earlier events on reported mood. In the fMRI experiment they found that the weights of the Primacy model correlated with prefrontal activation across subjects, while this was not the case for the Recency model.
The paper is clearly written and easy to understand. The question of how humans combine experienced events into reported mood is topical and the conclusions are striking, given the dominance of recency-based models in the literature (e.g., Kahneman's peak-end heuristic). The paper takes an interesting approach and presents an impressive amount of data.
However, at some points the arguments seemed a considerable stretch, in part because important experimental and methodological detail is missing, and in part because the analyses do not currently consider a number of potential confounds in both the models and the task design. Ultimately, these concerns come down to whether we can be certain that the results reflect a true primacy effect, as opposed to some other process that simply appears at face value to be a primacy effect. To this end, some important checks need to be made concerning both the computational and the fMRI analyses, as detailed below. These do require substantial extra modelling work, and it is quite possible that the conclusions will not survive these control analyses.
-
