Which of these is not included in the personal domain of social knowledge

Introduction

Thanks to advances in technology, massive amounts of information are now easily accessible. This includes personalized information about people’s past, present and future. Individuals must make many decisions regarding which information they would like to receive and which they would rather avoid. It is unclear how people make these choices.

Despite the relevance of this question to domains such as health, politics and science, we know surprisingly little about what drives information seeking. Nor do we have a clear understanding of why an individual decides to seek out particular information, while another actively avoids it. For example, a recent study1 found that approximately half of individuals surveyed wanted to know if they had a genetic predisposition to cancer, while the other half did not; half wanted to know the estimated global temperature in 2100, half did not; half wanted to know the amount of calories in meal options, half did not. Here, we characterize and quantify motives of information seeking and show how they explain individual differences in information-seeking choices.

We have recently proposed a theory which characterizes the key motives for information seeking2. According to this theory, when deciding whether to seek information, people first estimate what the information will reveal and then estimate the expected impact of that information on their action, affect and cognition. With regards to action, the prediction is that people want information more when it can aid in selecting action that will help gain rewards and avoid harm2. For example, people would be more likely to want to know about automobile safety ratings if they are about to buy a car, as the information can inform their purchasing decision. With regards to affect, all else being equal, people will be more likely to want information when they expect knowledge to make them feel better than ignorance [and vice versa]2,3,4,5,6,7,8,9,10. For example, the prediction is that a student would be more likely to want to know their mark on an exam if they believe they had done well. With regards to cognition, people will want information about concepts they think of often2. This is because such information is especially relevant to their internal representation of their world and highly connected with many other concepts2. For example, the prediction is2 that a person who thinks about dogs frequently, would be more interested in learning whether dogs are related to wolfs compared to someone who rarely thinks about dogs.

The estimated impact of information on action, affect and cognition is referred to as instrumental utility, hedonic utility and cognitive utility, respectively2. Each of these estimates can be positive [increasing information seeking], negative [increasing information avoidance] or zero [inducing indifference]2. We hypothesized that these estimates are integrated into a computation of the value of information, which will trigger information seeking or its active avoidance2. Here, over five studies testing 543 participants we provide an empirical test of this theory. To examine if the theory is domain general or domain specific, we test information seeking in three different domains—information about self-traits, finance and health.

We had further proposed that each of the three factors may be weighted differently, influencing the decision to seek or avoid information to different degrees2 [Fig. 1a]. Individual differences in information seeking may be related to the different weight individuals assign to each motive. For example, certain individuals may care most about the instrumental utility of information, whereas others may care most about the need to regulate their affective state, while other may assign equal weight to all three motives when seeking information, etc. Here, we quantify those differences and examine to what degree they are stable, or change, over time within and across domains, by conducting three longitudinal studies.

We had hypothesized that the weights people assign to each motive are related to self-reported mental health2. The reason for this hypothesis is that many psychopathology symptoms can be broadly characterized as problems in affective processes, cognitive functions as well as action planning and execution11,12. Abnormalities in these domains may reveal themselves in the type of information people choose to seek or avoid. For example, depression is characterized by a reduction in the belief that one has agency over outcomes13, which may lead to a reduction in the impact of instrumental utility on information seeking. As poor mental health is often associated with problems related to self-perception and thoughts regarding the self14,15,16,17,18,19,20, we test the relationship between mental health and information seeking in the domain of self-referential knowledge. If indeed psychopathology symptoms are related to specific patterns of information seeking, there is potential for using measured markers of information seeking to diagnose mental health problems.

Given this rich potential, it is surprising how limited our knowledge is of the links between mental health and information seeking. In fact, despite information seeking being central to human behaviour, we know remarkably little about how to quantify it or the mechanisms that underlie it. To address these unknowns, we conducted five studies in which participants were asked to indicate whether they would want to receive 40 pieces of information. In Experiment 1, 2 and 5 the information was related to self-traits, in Experiment 3 to finance and in Experiment 4 to health. Participants also provided ratings which served as proxies for the instrumental, hedonic and cognitive utility they assigned to each potential piece of information. These proxies were then used to quantify participants’ information-seeking motives and explain individual differences in participants’ choices. Experiment 1 and 3 were longitudinal studies that enabled us to quantify the stability of the motives over time within an individual and domain, and Experiment 4 examined stability over time across domains. Additionally, in Experiments 1 and 2 we assessed participants’ mental health using a battery of self-report psychopathology questionnaires21,22,23,24,25,26,27,28,29 and examined these responses for an association between mental health and information-seeking motives. In particular, we implemented a dimensionality approach30,31,32, which considers the possibility that a specific symptom is predictive of several psychiatric conditions, thus allowing an investigation that cuts through classic clinical boundaries.

Results

Task overview [Experiment 1]

Participants were asked to imagine that their family/friends had rated them on different attributes [for example, ‘intelligent’, ‘unreliable’]. In block one, on each of the 40 trials, participants indicated whether they would like to know how others had rated them on a specific attribute using a six-point Likert scale from −3[definitely don’t want to know] to +3[definitely want to know], with ‘0’ not included [Supplementary Fig. 2]. On average participants rated their desire to receive information as 0.43 [SD = 1.30], which is significantly different from the mid-point of the scale, t[79] = 2.970, p = 0.004.

In block two, participants provided the following ratings on a seven point Likert scale for each of the 40 traits: [i] their expectations regarding how useful it would be to know how others rated them on that trait [from −3 ‘not useful ‘ to +3 very useful], which provided an estimate of Instrumental Utility [e.g. how useful would it be to know how others rated you on ‘intelligence’?]; [ii] how they expect to feel if the rating was revealed to them [from −3 ‘very bad’ to +3 ‘very good’; e.g. how will you feel if you knew how others rated you on ‘intelligence’?] and how they expect to feel if the rating was never revealed to them [from −3 ‘very bad’ to +3 ‘very good’; e.g. how will you feel if you never knew how others rated you on ‘intelligence’?]. The difference between the last two ratings provided an estimate for Hedonic Utility and [iii] how often they think about each attribute [from −3 ‘never ‘to +3 ‘very often’; e.g. how often do you think about ‘intelligence’?], which provided an estimate of Cognitive Utility. The questions were selected based on the theory paper2 in which we had introduced the three utilities of information seeking and suggested quantifiable predictions. We note that these are not necessarily the only questions one can use to measure the three utilities, but we had proposed them as central ones in our original theory paper2.

Additionally, we asked participants to indicate how they expected others would rate them [from −3 ‘not at all this trait’ to +3 ‘very much this trait’]. This was done for two reasons. First, our theory suggests that people’s estimates of utilities are partially based on what they expect the information would reveal. For example, in order to estimate one’s affective response to information one needs to predict what the content of the information would be. Second, this question then allowed us to ask participants about their confidence in the above rating [−3 ‘not certain’ to +3 ‘very certain’]. That is, how confident [certain] they are of what information would reveal. Many studies suggest that uncertainty is related to information seeking33,34,35,36,37,38. Sometimes people want information about things they are certain about [a form of conformation bias36,37,38] and sometimes they want information about things they are uncertain about33,34,35, with one study suggesting that the sign of the effect can vary according to the environment39. Descriptive statistics of all these ratings and their inter-relationships are displayed in Supplementary Table 1.

Information seeking is best explained by taking into account instrumental, hedonic and cognitive utilities [Experiment 1]

We tested 99 participants on the information-seeking task described above. Eighty participants passed the attention check and had enough variability in their rating data to generate three beta coefficients [that is did not insert the same rating for all stimuli on any of the scales]. We submitted their data into a mixed-effects model to estimate the relationship between Instrumental Utility, Hedonic Utility and Cognitive Utility [which were estimated using the ratings as described above] and the desire to receive information [see methods]. Each of these three factors were centered within participant for each rating across all trials and included in the model as fixed and random effects. Random intercept and slope were estimated for each participant as well as random intercept for each item [see methods]. This revealed a significant fixed effect of Instrumental Utility [β = 0.114 ± 0.029 [SE], t[60.17] = 3.918, p = 0.001, Fig. 1b], Hedonic Utility [β = 0.123 ± 0.022 [SE], t[61.28] = 5.531, p = 0.0001, Fig. 1b] and Cognitive Utility [β = 0.091 ± 0.031 [SE], t[89.98] = 2.935, p = 0.004, Fig. 1b]. In particular, participants expressed a greater desire for knowledge when they believed the information would be useful, would have a more positive impact on their affect than ignorance, and also for stimuli they thought of frequently [see Supplementary Information for a study testing three additional motives of information seeking].

Fig. 1: Information-seeking motives.

a Information seeking and its avoidance is hypothesized to be driven by Instrumental Utility, Hedonic Utility and Cognitive Utility2. These values reflect the predicted impact of information on action, affect and cognition, respectively. These estimates are hypothesized to be integrated into a computation of the value of information, with different weights [β1–3] assigned to each of the three factors. The integrated value can lead to information seeking or avoidance. b Plotted are the beta coefficients from a linear mixed-effects model [N = 80 participants], showing that participants’ desire to receive information was greater when the Instrumental Utility [p = 0.001, two sided], Hedonic Utility [p = 0.0001, two sided] and Cognitive Utility [p = 0.004, two sided] of information were higher. These were estimated respectively by participants’ ratings of how useful the information would be, how they would feel to know vs not to know, and how frequently they think about the stimulus. The horizontal lines indicate median values, boxes indicate 25–75% interquartile range and whiskers indicate 1.5 × interquartile range; individual scores are shown as dots. c BIC scores reveal that the model described in b fit the data better than models including alternate combinations of the utilities and also those including participants’ confidence regarding what the information would reveal. The same was true when examining AIC scores [see Supplementary Table 8]. Smaller BIC and AIC scores indicate better fit. d Plotted are the weights each individual put on each motive when seeking information. Beta coefficients of Instrumental Utility are on the x-axis, of Cognitive Utility on the y-axis and of Hedonic Utility on the z-axis. Green dots represent participants who put the largest weight on Instrumental Utility when seeking information. Red dots represent participants who put the largest weight on Hedonic Utility when seeking information. Blue dots represent participants who put the largest weight on Cognitive Utility when seeking information. The colour gradient represents how dominant the largest weight was in comparison to the other two weights. Individuals who put more than twice as much weight on their dominant utility than the other two utilities are represented in darkest colours. Those whose dominant utility was less than 1.25 times larger than the other two are represented in the lightest colours. ***P  0.188]. These results suggest that the weight participants’ assign to Cognitive Utility, but not the other two utilities, when seeking self-referntial information is related to their mental health across the three-psychopathology dimensions, with greater weight on Cognitive Utility associate with better mental health.

To illustrate this result in a more simplified manner, we conducted a linear regression with mental health as the dependent measure [quantified as the average psychopathology score across the three dimensions] and the following predictors: the weight assigned to Instrumental Utility [β1] when seeking information, as well as that assigned to Hedonic Utility [β2] and to Cognitive Utility [β3]. Age and gender were also included as predictors. Confirming the analysis above, a significant inverse relationship was observed between mental health and the weight assigned to Cognitive Utility when seeking information [β = −1.053, p = 0.016], suggesting that participants who seek information more on issues they think of often are the ones who report less psychopathology symptoms across the board. No other predictor was significant [Instrumental Utility: β = −0.710, p = 0.094; Hedonic Utility: β = −0.072, p = 0.870; Age: β = −0.010, p = 0.893; Gender: β = −0.211, p = 0.296; Fig. 2b]. Finally, correlating each beta with the average psychopathology score across participants [controlling for age and gender], again reveals a significant association with the weight assigned to Cognitive Utility when seeking information [r = −0.244 [67] p = 0.043], but not with the weight assigned to Instrumental [r = −0.136 [67] p = 0.264] or Hedonic [r = 0.09 [67], p = 0.463] utilities.

Fig. 2: Information seeking related to psychopathology.

a Plotted are the weights [based on ref. 30] given to each questionnaire item21,22,23,24,25,26,27,28,29 when calculating the weighted score for each participant on each of the three psychopathology dimensions identified previously [“Anxious-Depression”, “Social-Withdrawal” and “Compulsive-Behaviour and Intrusive Thought”]. b Plotted on the y-axis is the average psychopathology score across the three dimensions described in a, Z-scored. On the x-axis are the weights assigned to each information-seeking motive from a linear regression predicting information seeking from Instrumental Utility [green], Hedonic Utility [red] and Cognitive Utility [blue]. Dots represent individual participants. Shading represents confidence interval. Line represents the relationship between the abscissa and ordinate controlling for the effect of the other two motives as well as of age and gender. As can be observed, participants who placed a large positive weight on Cognitive Utility when seeking information reported less psychopathology symptoms [p = 0.016, two sided], while we observed no effect of Instrumental Utility [p = 0.094, two sided] or Hedonic Utility [p = 0.870, two sided]. Error bars SEM. *P