# Predicting Mood Based on the Social Context Measured Through the Experience Sampling Method, Digital Phenotyping, and Social Networks Cite: Langener, A.M., Bringmann, L.F., Kas, M.J. _et al._ Predicting Mood Based on the Social Context Measured Through the Experience Sampling Method, Digital Phenotyping, and Social Networks. _Adm Policy Ment Health_ (2024). https://doi.org/10.1007/s10488-023-01328-0 > [!Info] > Relevance > For study design in the future Paper [[Predicting Mood Based on the Social Context Measured Through the Experience Sampling Method, Digital Phenotyping, and Social Networks.pdf]] [[Digital Phenotyping]] [[Mood EMA Index]] [[Social]] - Can we predict mood based on EMA /+ DP /+ Egocentric Network? ## Social Context is made up by 1. Social Experience includes daily social interactions, also known as personal social network 2. The perception of these interactions - whether pleasant or unpleasant - this relates the psychological/cognitive representations. ## How do they measure? 1. Through [[Digital Phenotyping]] 2. Through [[Ecological Momentary Assessments]] - Experience Sampling Method 3. Egocentric Networks 1. Friends, contacts. Identity. The roles, and closeness to those contacts. ## Result - Overall prediction accuracy is low. Similar to this study [[Asselbergs et al 2016]], ## Discussion - This study [[LiKamWa et al 2013]] have high prediction accuracy 93%. How come? - Maybe because they had longer period of study (60days), compare to 28 days in this paper and (42 days in [[Asselbergs et al 2016]]) # Different parts of the social context are important to predict moods - **Different people have different social contexts, and as such, we may need different methods to capture that**. - "*Our results suggest that the optimal set of social context predictors varied among participants and that there is variation among participants in which specific variable was most important for predicting mood. This makes sense, considering that different parts of the social context may be more or less important for predicting the mood of different individuals. For instance, for people with few social interactions throughout the whole study period, passive measures, such as app usage, might be a better indicator of their mood than the total minutes they spent in interactions.*" - **Phone calls itself not important, because many calls are now done through apps which is not recorded.** - "*We calculated Shapley values and found that phone calls recorded via Behapp are relatively unimportant for predicting mood. This is somewhat at odds with the finding that, for some participants (e.g., Participants 1 and 4), the total minutes spent calling recorded via ESM, is more important. This may be explained by the fact that Behapp only captures phone calls via the mobile network and no other forms of communication like (video) calls made via WhatsApp. Results support the integration of digital data sources to measure distinct aspects of social behavior that may otherwise be missed by relying solely on a single method.*" ## Limitations - ESM is not validated - [[Need to consider validity and reliability of questionnaires used]] - [[Construct Validity]] - We need to make sure that we are measuring what we are measuring. - "*a recent study suggests that researchers should be more critical when using a score that consists of multiple items and that using single items might be superior (Cloos et al., 2023; McNeish & Wolf, 2020). Thus, it would be interesting to investigate how results would change if we would only aim to predict the score of a single item, such as “I feel happy” or “I feel sad”, instead of positive or negative affect. In addition, the selection of items to measure positive and negative affect or other constructs in ESM research is often arbitrary. This can reduce construct validity (Bringmann et al., 2022; Flake & Fried, 2020), which could potentially lead to lower predictive performance when attempting to predict these constructs.*" - Different studies have different EMA which may have different construct validity. They also have different time frequency in their measurements, which makes it different to compare these studies. - "*The variability in predictive ability and importance of variables in predicting mood across participants indicates that it is difficult to generalize our findings, even in this sample with participants from fairly homogeneous backgrounds. Our findings also suggest that, despite our intensive data collection, the sample sizes for the prediction models may have been a limiting factor Thus, future research using individual-level tailored models to predict mood within individuals based using similar methods and larger sample sizes are recommended.*" # How do they check that their algorithm is accurate or not? They predict the mood to what the participants reported. # How do they create a baseline model? - They use the average mood score to predict the mood score for the next time interval. - Their advance model need to be more accurate then the base model to be closer to what participants reported. # Robustness Check, that affect result - Different phone may have different data quality, - How do we manage missing data?