The second main result was that the discrepancy between actual and desired IOS Scale closeness was greatest for those pairs with a preoccupied partner (Ms = -.70 for secure pairs; -.45, avoidant/dismissive; -.57, avoidant/fearful; -1.12, preoccupied with other; contrast p < .01). (The same significant pattern was observed considering only the scores on desired IOS Scale closeness, and there were no significant or near-significant differences between the score of the preoccupied partner vs. the nonpreoccupied partner.) Combined with the results for actual closeness, the overall pattern is one in which pairs with a preoccupied partner report about the same level of closeness as other pairs but are considerably less satisfied with that level of closeness. (This result is consistent with the attachment theory description of preoccupied individuals as wishing for more closeness than they are able to find; e.g., Hazan 8c Shaver, 1987.) Once again, these results illustrate the usefulness of the closeness-generating procedure; without it, it would be quite difficult to test this kind of issue.
The third main result was about change in reported attachment style from before to after the task. To simplify the analysis and maximize interpretability, we combined the change on the four attachment-style scales into two uncorrelated linear composites corresponding to Bartholomew and Horowitz's (1991) scheme: (a) model of self (increases in secure and avoidant/dismissive minus increases in preoccupied and avoidant/fearful) and (b) model of other (increases in secure and preoccupied minus increases in avoidant/dismissive and avoidant/ fearful). (Results using change on individual attachment styles were entirely consistent with the composite results.) Median test-re test correlation for pairs for model of self was .77 and for model of other, .76.
Overall, from before to after the task, there was little change in model of self (M change = -.10), with increased positive ratings for the avoidant/fearful and preoccupied pairings offsetting decreases for the other two groups—a pattern of differences that may simply reflect regression to the mean. However, there was a clear and significant overall increase on positivity of model of other (Af change = 1.11); F(l, 93) = 30.19, p < .01. (The amount of increase was significantly greater for avoidant/dismissive than for secure pairs as might be expected from regression to the mean. But all four groupings showed an increase of some degree, which would not be expected from regression to the mean.) This increase was significantly greater (p < .01) than the overall change for self-model, and this difference was not qualified by an interaction with attachment-style pairing. If one interprets these changes as actual modifications of one's mental model of other, these various findings are consistent with Hazan and Shaver's (1987) suggestion that relational experience can have an impact on attachment style. On the other hand, the entire tenor of attachment theory emphasizes that mental models are formed early and are not easily modified by later experience. Perhaps one way of understanding the present data is in terms of a temporary modification, a kind of tempering, of the degree of extremity of one's model of other from an experience with such an impact that it is counter to what one expects from one's model. But even if changes produced by this procedure are only temporary or shallow, they may well represent in a small way the kind of processes that actually modify attachment style over the longer term. To the extent this may be the case, these data illustrate the potential of the closeness- generating procedure for being able to model in the laboratory processes that are otherwise not subject to high levels of experimental control.
STUDY3
Study 3 examined the effect of making closeness an explicit task. In addition, Study 3 again attempted to illustrate the usefulness of our procedure for evaluating theoretical issues by considering a different area than attachment style—introversion/extraversion. Using this issue for illustration demonstrates the usefulness of our procedure for controlling the matching of subjects into pairs using a different individual difference variable than attachment and also demonstrates how an experimental manipulation of the instructions may create a theoretically interesting interaction effect.
The impact of making closeness an explicit task. In Studies 1 and 2 and in our preliminary research with this procedure (Aron, Aron, Melinat, & Vallone, 1991), subjects were explicitly instructed to make closeness a goal. As noted in the overall introduction, we intentionally adopted this approach to enhance the degree of closeness achieved, and we considered this approach similar to procedures used in the study of perception, memory, and other areas of experimental psychology. Study 3 directly examined the impact of this feature of our procedure by randomly assigning half the pairs to an experimental condition in which there was no indication before or during the study that closeness was the goal.
This manipulation was primarily intended to help determine whether this aspect of our procedure contributes to the overall closeness achieved. This manipulation was also important for considering the possible role in our procedure of demand characteristics (i.e., effects of subjects' knowledge of the purpose of the study). In general, we do not think demand characteristics are a great problem in using our procedure in research because the procedure is intended for studies that examine not whether closeness develops overall but the amount of it that develops differentially across experimental conditions. Any effect of demand characteristics seems to be equated across experimental conditions. However, the possible role of demand characteristics did seem potentially relevant to the more general issues of just how much "real" closeness is being produced by these tasks. Thus we felt it would be useful to examine whether there was a substantial degree of closeness attained even when efforts were made to see that subjects did not know in advance or during the procedures that the study had to do with closeness.
Extraversion/introversion. Extraversion/introversion appears to be a consistent factor in the latent structure of ordinary persons' ratings of self and others (e.g., Goldberg, 1990). Extraversion/in troversion scales (e.g., Eysenck 8c Eysenck, 1975; Myers 8c McCaulley, 1985) emphasize the respondent's desire and ability to socialize with strangers. However, it is not entirely clear how introverts and extraverts behave in situations perceived as offering a possibility for real intimacy. On one hand, extraverts would be expected to be most comfortable and effective in any situation with a new other. On the other hand, the potential for one-on-one intimacy is consistent with the kinds of relations that are particularly desirable for introverts (e.g., Hotard, McFatter, McWhirter, & Stegall, 1989). However, this issue is not easy to sort out in correlational research. Thus Study 3 created pairings of extraverts with extraverts, introverts with introverts, and mixed-type pairs and examined how close each pairing became—illustrating the usefulness of our procedure for independently controlling with whom people have the opportunity to develop closeness. Study 3 also permitted us to examine whether there were any differential effects on closeness of these pairings as a function of whether closeness was an explicit goal— illustrating the usefulness of our procedure for manipulating a situational factor (closeness as a goal of the situation) relevant to how closeness may develop. The extraversion/introversion aspect of Study 3 was suggested by a somewhat similar study to ours conducted by Thorne (1987) in which women paired by introversion/ extraversion participated in 10-min unstructured conversations and then made ratings of each other's personality.5