Metacognition – Cognitive and Social Dimensions (II)

Here’s a link to my first post about the book. I decided that it would be too much work to cover the rest of the book in one post (especially as it’s really bothersome to blog this in the first place, because I as mentioned was unable to highlight while I was reading it), so you can expect a third post about the book at some point in the future.

I know there’s been a lot of psychology-related stuff on the blog lately, but I’m currently finishing a math textbook and I’ll most likely finish a book on biological aspects of aging tomorrow, so you can probably expect me to cover some other topics as well in the near future.

Below I have added some quotes and some comments/observations of my own.

…

“In accordance with the general notion of Festinger’s social comparison theory (1954), we assume that individuals are particularly susceptible to social influence if they are not confident in their own judgments of the situation. In order to reduce their uncertainty, individuals will rely on information that is, intentionally or unintentionally, subtly or obviously, provided by others. Shifting these considerations from attitudes, beliefs, or behaviors to memory, the belief that an event has occurred can be treated like an attitude that a person may hold. Social influence on memory should then be most likely to occur when individuals lack confidence in their memorial beliefs […] memory is susceptible to social influence if individuals are not confident that the absence of a recollection experience implies the non-occurrence of an event, and if individuals are not confident where the presence of a recollection experience comes from. Thus, it is essential to identify variables that increase or decrease individuals’ confidence in their memory experiences, and to test whether social influence on memory is a function of these variables. […] Whether or not the lack of a recollective experience is sufficient to infer the non-occurrence of an event may often depend on whether the event is perceptually or categorically distinct. […] Specifically, if the item is held to be highly memorable, individuals should be very confident that the absence of a recollective experience implies non-occurrence. In this case, individuals are unlikely to rely on information provided by others, and should therefore be minimally susceptible to social influence. Individuals face a different situation, however, if they cannot remember an event and have no reason to believe that the event in question is particularly memorable. In this case, they remain uncertain whether the absence of a recollective experience is due to the non-occurrence of the event or due to a failure of remembering the event. […] we may speculate on general conditions that increase individuals’ confidence in their own attitudes, beliefs, behaviors, or recollective experiences, and consequently decrease the susceptibility to social influence. In our view, individuals’ self-esteem could constitute a prime candidate for such a general impact. Particularly, we speculate that individuals with chronically or temporarily low self-esteem should be less likely to rely on their own subjective experiences and instead rely on subtle cues in the environment. In contrast, individuals with chronically or temporarily high self-esteem should be more confident in their own subjective experiences. As a consequence, individuals with low self-esteem are expected to be more susceptible to social influence.”

This last part in particular is somewhat speculative and they don’t have any studies to point to, but that’s because research into those topics had not (yet?) been done at the time the book was written which is hardly the fault of the authors. I’m actually somewhat curious if people have looked into this kind of stuff since then. It’s quite plausible that variables such as these play a role in memory formation.

…

“we all have a powerful intuition that we know our relationship partners. In fact, when it comes to lovers, this feeling of mutual understanding and “knowing” is so powerful that we have invented phrases like “soul mate” and “other half” to describe relationships characterized by it. Despite the pervasive feeling that we come to know one another better as relationships progress from casual to intimate, there is little evidence to suggest that this feeling is based on true gains in accuracy. For example, longitudinal studies have shown that if accuracy increases over time at all, after the initial phases of the relationship such gains are minimal (for a review, see Kenny, 1994). […] The lack of clear evidence that people’s perceptions grow increasingly accurate as relationships progress is so surprising because such gains in accuracy seem self-evident to most of us.”

“We suggest that as people integrate information about their relationship partners into coherent impressions, their representations of those persons become “richer.” One way to increase representational richness is by adding novel information. The longer we are acquainted with people, for example, the more information we gather about them and the richer our representations of them become. Representational richness will also increase as our impressions become better integrated and more coherent. As we become more deeply involved with someone, we may carefully organize and integrate what we know about them even though individual elements may be incongruous with one another […]. Whatever the source of increments in representational richness may be, richness will foster accessibility […] and accessibility will, in turn, promote confidence […]. Although representational richness should make people more confident of their perceptions, it will not necessarily contribute to the accuracy of their perceptions. The reason is that any information may increase representational richness but only information that is truly diagnostic will foster accuracy (Funder, 1995). Moreover, there is ample evidence that people use non-diagnostic information as a basis for forming impressions of others. For example, people are inclined to infer dispositions from behaviors that are, in reality, constrained by situational factors […], attribute characteristics based on dubious “implicit theories” of personality […], and fail to revise their initial opinions of others even when the evidence suggests they should […]. This tendency to make inferences from non-diagnostic information (e.g. category membership, situationally produced behavior, implicit theories of personality), in conjunction with a tendency to use such information as a basis for increasing confidence, could result in people being highly confident of beliefs that are inaccurate and misleading. To test this conceptualization of confidence and accuracy, we have conducted a series of field and laboratory studies.”

“The first study was an investigation of 57 heterosexual couples aged from 17 to 41 (Swann & Gill, 1997). Participants had been dating from three weeks to just over six years, with an average of one and one-half years. […] The task of targets was to rate themselves on a series of four questionnaires, including the Sexual History Questionnaire (SHQ; which includes 10 open-ended questions concerning the respondents’ past sexual behavior), the Self-Liking/Competence Scale (SLC; a measure of global self-esteem), the Self-Attributes Questionnaire (SAQ; a measure of self-perceived intelligence, social competence, artistic/musical ability, athleticism, and physical attractiveness), and the Activity Preferences Questionnaire (APQ; a measure of enjoyment of 37 leisure activities). While targets rated themselves, perceivers guessed how targets would rate themselves and indicated their confidence that their guesses were correct. […] We tapped representational richness in two ways. First, we measured time in the relationship as a proxy for amount of information [this seems like a plausible strategy, but obviously not without problems – US]. Second, we measured involvement in the relationship as a proxy for motivation to integrate information. […] Our major prediction was that representational richness would increase confidence but not necessarily accuracy. Regression analyses confirmed that relationship length and involvement were both associated with confidence but had virtually no relation to the accuracy of perceivers’ knowledge of their partners’ self-ratings. […] confidence ran high while accuracy remained rather modest.”

They did another similar study on 40 university students from Texas. That one dealt with roommates, not romantic partners:

“During the first week of the semester, participants rated themselves and attempted to predict the self-ratings of their roommates. […] [One difference was that] we replaced the SHQ with the SAQ-R, a measure that focused on 10 personality characteristics (sense of humor, extroversion, assertiveness, etc.) […] We collected these ratings during the first week of the semester and again six weeks later. During the second session participants also answered four questions about their level of involvement in the relationship with their roommate (how much time they spent doing things with their roommate, how many conversations they had, how often these conversations dealt with relatively private issues, and how much they liked their roommate). […] For the most part, time was unrelated to accuracy, although there was a slight tendency for accuracy on the SAQ to improve over time. Similarly, there was virtually no relation between involvement and accuracy on the SAQ-R or the SLC and a slight negative relation between involvement and accuracy on the SAQ and the APQ. On the latter two measures, then, involvement was associated with increased confidence but decreased accuracy. […] people [seem to be]
sensitive to the accuracy of their knowledge of their partners’ self-esteem but little else. […] Taken together, the results of our studies of dating couples and roommates provide converging evidence that relationship length and involvement foster confidence. Furthermore, the link between time and confidence does not seem to reflect artifactual processes such as a tendency for people who are unconfident to end their relationships. Finally, the fact that relationship length and involvement were associated with increases in confidence but had little impact on accuracy supports our suggestion that confidence grows out of processes that are unrelated to accuracy.”

…

I’m not sure how convinced I am at this point that the above findings are ‘true’ (‘how to interpret the findings’), but they are certainly thought-provoking. I’ve long thought of ‘convergence’ as the default consequence of long-term social exposure/bonding, but I can’t really remember where I picked up that idea (maybe Pamela Regan’s book?), and on the other hand it’s also obvious that you need stuff besides convergence to explain things like divorce-statistics and so the process is far from independent of other relevant relationship variables. Funder talked about Kenny’s results as well so this stuff is not completely new to me, but I still find it slightly weird to think about. They note in the text that although there was naturally some variation in the performance, there were no people in their studies who were consistently ‘right’; if people were right on one dimension, they tended to be wrong on another dimension. A thought I remember having while reading the book was that important variation might be overlooked due to the way the variables were applied in their analysis; in particular, I thought it might have been useful to get a measure of how salient the individual variables included in the various constructs were considered to be by the targets (the people who were judged). It seems to me that one should expect there to be important individual-level differences in terms of which variables are considered to be most ‘important’ by the people who get evaluated; not all variables contribute the same to the personality assessment of an individual, as some people will be more or less defined by their high level of extroversion, others by their conscientiousness, etc., and implicitly assuming that all variables included in a composite metric contributes the same to the understanding of how an individual ‘works’, which seems to be what the researchers did here, may be problematic. A weighting procedure of some sort might have been very useful. The confidence estimates relate only to the level of certainty, not the importance/salience of the variable. Obtaining ‘validated’ weightings provided by the target him-/herself, perhaps constructed using target responses from personality inventories, might have been a good idea. There would be some problems associated with interpreting such a variable, as mentioned below, but I think it would have added relevant information.

One might argue that if people were really great at figuring out what other people are like (‘the long-term partners got all the ‘salient variables’ right, whereas the short-term partners didn’t’), the analysis should not make them out to be systematically overconfident. There are however a few important details to have in mind when thinking about this overconfidence. One thing that’s important to note in this context is the way the criterion problem is handled; in the specific setup these studies apply it’s actually far from trivial to go from ‘disagreement between judge and target about a variable’ to ‘the judge was wrong (and (/thus) overconfident)’. Another interpretation might be that the target was overestimating (or, if you want to put on a more cynical hat, lying about…) e.g. how confident or assertive he/she really is, and that the judge’s evaluation was actually closer to ‘the truth’ than that of the target (for much more on such factors and the role they play in personality judgment, I refer again to Funder), and that the increased confidence on part of the long-term partner was really justified because she really isn’t that assertive, even if she claims to be. You might mix this stuff with some observations related to e.g. how social feedback processes vary over time in relationships (e.g., ‘a new romantic partner will be much less confident because getting an important answer/estimate wrong may end the relationship, whereas the partner in an established relationship risks less and so may be more confident about his/her partner-relevant judgments’); I’m not quite sure where you’ll end up, but there’s a lot of stuff that could potentially be going on here. If such dynamics are driving the results to some extent, the relevant conclusion may perhaps not be that ‘people are bad at figuring out stuff about others’ and that ‘people get overconfident when forming impressions about how good they are at figuring out what another individual is like when they’ve known that individual for a long time’, but rather that ‘people are bad at figuring out stuff about themselves, and hold views about themselves which are inaccurate’. This they surely do, we know that people’s self-perceptions are often at variance with the truth to some extent and that part of that variance is systematic and can be formalized – I’ve talked about such stuff many times before here on the blog, so I don’t see any reason to repeat myself – but the extent to which such aspects may impact the specific results and their interpretation is not perfectly clear to me at this point.

One may well argue that the overconfidence element is not so easy to get rid of, in the sense that regardless of whether the judgments of the judges were ‘actually true’ (‘X isn’t actually that assertive, even though he claims to be’), it’s clear that the judges did not ‘get better over time’ (given the applied identification procedure is valid) at figuring out how the targets would present themselves and answer the questions they were asked, and that couples who’d been together longer were more confident yet did not get enough answers right to justify that increased level of confidence. I’d also not disagree with the notion that it may well be questionable how much a model explaining the increased divergence between confidence and accuracy as a result of accurate beliefs on part of the judge combined with ‘faulty’ self-concepts on part of the target really adds, because self-concepts are a very important part of the identity of an individual; ‘if you really know her that well, you should know that she prefers to think of herself as being assertive’. Regardless of how you want to interpret the combination of an increased confidence ‘over time’ and a stagnant hit rate, something important is clearly going on here – but the underlying dynamics do not to me seem at all well-elucidated. One approach they do not, as far as I can remember, talk about but which sort of seems logical to apply given the context of the book, is perhaps also to think in terms of explanations invoking societal norms and implicit theories to explain the confidence/accuracy gap; one could argue that the people who had known each other longer in the study didn’t actually think they knew the answers to the questions posed much better than the people who had only known each other for a short amount of time, but they did know that other people believe they ought to know the answers (‘when people have known each other for a long time, they’re supposed to know that sort of stuff’) – this would be one approach which to me seems consistent with the ‘confidence and accuracy are independent variables’-view which takes into account potential relationship-length-related shifts in response strategies.

All those things said, it should be noted that the findings don’t just stand on their own as isolated yet interesting observations, and that the observed overconfidence may be ‘real’ and not the result of unaccounted-for methodological problems – as they argue in their conclusion to the chapter:

“Our evidence of dissociations between confidence and accuracy fits nicely into a growing body of literature suggesting that subjective indicators of knowing are often unreliable indicators of objective knowledge or comprehension (see Jacoby, Bjork, & Kelley, 1994, for a review). For example, people misjudge their comprehension of texts […], the correctness of their answers to general knowledge questions […], the correctness of recalled letter strings […], and the correctness of their eyewitness identifications […]. Reder and Ritter (1992) have even shown that familiarity with a question predicts people’s confidence that they know the answer better than does familiarity with the answer. Taken as a whole, this literature suggests that what people think they know is not always what they really know.”