Cognitive Vulnerability to Depression Can Be Contagious

Method

Results

We tested two primary hypotheses. First, we predicted that cognitive vulnerability to depression would be contagious between roommates. Second, we predicted that changes in cognitive vulnerability would lead to changes in risk for future depressive symptoms. As expected due to random assignment, there was no association between roommates’ levels of cognitive vulnerability at baseline: The intraclass correlation for cognitive vulnerability as measured by the RRS was –.04, p = .66; the intraclass correlation for cognitive vulnerability as measured by the CSQ was –.06, p = .73.

Cognitive vulnerability to depression is contagious

We used the actor-partner interdependence model (APIM; Kenny, 1996) to test whether cognitive vulnerability was contagious between roommates. Specifically, we used multilevel modeling for indistinguishable dyads to test the stability of cognitive vulnerability over time and determine how a participant’s vulnerability level was influenced by his or her roommate’s vulnerability level. APIM is superior to traditional statistical strategies for testing interpersonal phenomena for a number of reasons. First, unlike traditional statistical techniques, APIM does not assume independence of observations. This is important because the assumption of independence is often violated in studies of interpersonal relationships; this violation can then lead to biased statistical significance tests (Cook & Kenny, 2005). Second, APIM accounts for bidirectional effects; it estimates both the “actor’s” effect on the “partner” as well as the “partner’s” effect on the “actor.”

In our analyses, the dependent variable was cognitive vulnerability (as measured by the RRS and CSQ, respectively) at 3 months and 6 months. We controlled for baseline levels of depressive symptoms (BDI) and recent stressful life events (ALEQ) to ensure the results could not be attributed to any individual differences in negative mood or the shared experience of stressful life events. The independent variable was level of cognitive vulnerability at baseline (as measured by the RRS and CSQ, respectively).

Consistent with prior research, results showed that cognitive vulnerability was relatively stable across the prospective interval. The test-retest correlations for both cognitive vulnerability measures ranged between .5 and .7 for the prospective interval. Participants’ level of cognitive vulnerability at baseline was a robust predictor of their level of cognitive vulnerability at both 3 months and 6 months (see Figs. 1 and 2). Of importance, the results supported our main hypothesis that cognitive vulnerability would be contagious between college roommates. As shown in Figure 1, a person’s future level of cognitive vulnerability was significantly affected by his or her roommate’s baseline level of cognitive vulnerability (and vice versa) as measured by the RRS. Specifically, individuals who were randomly assigned to a roommate with high initial levels of cognitive vulnerability experienced increases in their own level of cognitive vulnerability over time, whereas those who were assigned to a roommate with low initial levels of cognitive vulnerability experienced decreases in their levels of cognitive vulnerability. This contagion effect was significant for the RRS at both the 3-month, b = .17, t(155) = 2.34, p = .02, and 6-month intervals, b = .16, t(132) = 2.13, p = .04, even after controlling for baseline levels of cognitive vulnerability, depressive symptoms, and stressful life events for both members of the roommate pair (note that the results also hold if roommates’ level of depressive symptoms at follow-up [3 months and 6 months, respectively] is also controlled for). The contagion effect was not significant when cognitive vulnerability was measured with the CSQ. As shown in Figure 2, the cognitive vulnerability as measured by the CSQ was more stable across time, and there was not a significant contagion effect at 3 months, b = .01, t(154) = 0.07, p = .94, or at 6 months, b = .12, t(133) = 1.83, p = .07. It is important to note that neither roommates’ levels of depressive symptoms nor their levels of stress was a significant predictor of changes in participants’ levels of cognitive vulnerability (either the CSQ or RRS) at either 3 months or 6 months.

OPEN IN VIEWER

Fig. 1.

Results of the actor-partner interdependence model testing whether cognitive vulnerability, as measured by the Ruminative Response Scale (RRS), was contagious between randomly assigned roommates at 3 months (top panel) and 6 months (bottom panel) controlling for depressive symptoms and negative life events (for both members of the roommate pair).

OPEN IN VIEWER

Fig. 2.

Results of the actor-partner interdependence model testing whether cognitive vulnerability, as measured by the Cognitive Style Questionnaire (CSQ), was contagious between randomly assigned roommates at 3 months (top panel) and 6 months (bottom panel) controlling for depressive symptoms and negative life events (for both members of the roommate).

Changes in cognitive vulnerability predict future depressive symptoms

We used a hierarchical multiple regression analysis (Cohen, Cohen, West, & Aiken, 2003) to test whether increases in cognitive vulnerability from baseline to 3 months later increased risk for depressive symptoms at the 6-month assessment under conditions of stress. The dependent variable was level of depressive symptoms at the 6-month assessment. Covariates (depressive symptoms at baseline and stressful life events at 3 months) were entered into the first step of the regression equation. In the second step, the main effects of stressful life events (ALEQ at 6 months) and changes in cognitive vulnerability (e.g., RRS change score from baseline to 3 months) were entered. Last, the Vulnerability × Stress interaction term was entered (RRS change score × ALEQ at 6 months). Consistent with hypotheses, there was a significant RRS × ALEQ interaction, b = –.07, t = −2.00, pr = −.17, p = .048, (model R² = .46). As shown in Figure 3, participants who experienced an increase in cognitive vulnerability during the first 3 months of college exhibited significantly greater levels of depressive symptoms at 6 months than did participants who did not experience an increase in cognitive vulnerability, particularly under conditions of high stress.

OPEN IN VIEWER

Fig. 3.

Depressive symptoms at 6 months as a function of change in cognitive vulnerability (increase vs. no increase in RRS score from baseline to 3 months) and stress (low vs. high).

The results of the regression analysis supported our hypothesis; however, a limitation of this analysis was that we could not determine if the portion of change in cognitive vulnerability due to contagion was what predicted future depression. To address this limitation, we conducted a complementary analysis using the APIM. If the contagion effect is responsible for changes in risk for depression, then changes in cognitive vulnerability should mediate the association between roommates’ level of cognitive vulnerability and participants’ future depressive symptoms. In other words, roommates’ level of cognitive vulnerability at baseline should predict changes in participants’ future level of depression because of its effect on participants’ own level of cognitive vulnerability. Following the criteria for mediation set forth by Baron and Kenny (1986), the following four conditions should hold if mediation is present: (a) Roommates’ baseline level of cognitive vulnerability should predict participants’ future level of depressive symptoms, (b) roommates’ baseline level of cognitive vulnerability should predict changes in participants’ level of cognitive vulnerability, (c) changes in participants’ level of cognitive vulnerability should predict future depressive symptoms when controlling for roommates’ baseline level of cognitive vulnerability, and (d) roommates’ baseline level of cognitive vulnerability in the third condition should no longer be a significant predictor when controlling for changes in cognitive vulnerability. Note that Condition b was already supported in the earlier analysis; thus, these analyses focused on Conditions a, c, and d.

First we tested if roommates’ level of cognitive vulnerability (RRS) at baseline predicted participants’ level of depressive symptoms at 6 months (Condition a). As hypothesized, the APIM analyses found that individuals’ future levels of depressive symptoms were directly influenced by their roommates’ baseline level of cognitive vulnerability. Participants who were randomly assigned to a roommate with high initial levels of cognitive vulnerability experienced greater levels of depressives symptoms over time than those assigned to a roommate with low initial levels of cognitive vulnerability experienced (even after controlling for baseline levels of cognitive vulnerability, depressive symptoms, and stressful life events for both members of the roommate pair), b = .32, t(124) = 2.36, p = .02. Next, we tested for the mediation effect by adding RRS change score (i.e., the mediator) to the APIM to determine if it reduced the power of roommates’ level of cognitive vulnerability to predict participants’ future depressive symptoms (Conditions c and d). Results confirmed the mediating effect of changes in cognitive vulnerability (Sobel’s t test p = .054). When controlling for changes in participants’ level of cognitive vulnerability from baseline to 3 months, roommates’ baseline level of cognitive vulnerability was no longer a significant predictor of participants’ depressive symptoms at 6 months. Changes in cognitive vulnerability (i.e., the mediator) remained a significant predictor of depressive symptoms, b = .35, t(123) = 2.23, p = .03.

It is important to note that roommates’ level of depressive symptoms was not a significant predictor of changes in participants’ level of depressive symptoms at either 3 months (p > .40) or 6 months (p > .40). In other words, there was not a contagion effect for depressive symptoms.

Discussion

The results of the study support the hypothesis that cognitive vulnerability to depression can be contagious. We found that participants’ level of cognitive vulnerability was significantly influenced by their roommates’ level of cognitive vulnerability and vice versa. Participants who were randomly assigned to a roommate with high levels of cognitive vulnerability (specifically a ruminative response style) were likely to “catch” their roommate’s cognitive style and develop higher levels of cognitive vulnerability over the prospective interval. This contagion effect was detectable after only 3 months of cohabitation. It is important to emphasize that these results cannot be explained by participants’ and roommates’ levels of depressive symptoms or by the occurrence of stressful life events (neither roommates’ level of depressive symptoms nor their level of stress was a significant predictor of changes in cognitive vulnerability). Not only was cognitive vulnerability contagious, but changes in vulnerability affected risk for future depressive symptoms. Participants who experienced an increase in cognitive vulnerability levels during the first 3 months of college had nearly twice the level of depressive symptoms at 6 months than did participants who did not experience an increase in cognitive vulnerability, particularly under conditions of high stress.

To date, there has been little reason to view cognitive vulnerability as anything other than highly stable, akin to a genetic diathesis. However, the results of this study indicate that it may be time to rethink how cognitive vulnerability is conceptualized. Our study demonstrates that cognitive vulnerability has the potential to wax and wane over time depending on the social context. This means that cognitive vulnerability should be thought of as plastic rather than immutable. Our findings are consistent with a growing number of studies that have found that many psychological and biological factors previously thought to be “set in stone” by adulthood continue to be malleable. For example, it was assumed for nearly a century that the central nervous system became structurally stable in early childhood and that the brain was not capable of generating new neurons in adulthood. But there is now strong support for neurogenesis in adult mammals (Gould & Gross, 2002). In addition, it was once believed that a second language could not be readily acquired after a critical period in childhood. However, we now know that given the appropriate environmental context (e.g., immersion), adults can very successfully achieve proficiency in a second language (e.g., Flege, Yeni Komshian, & Liu, 1999). Along these lines, we found that cognitive vulnerability can still be altered by the social environment (via contagion) years after its consolidation in early adolescence. Our finding opens the door for an entirely new line of research investigating how cognitive vulnerability changes over the life span, the mechanisms by which change occurs, and their effect on developmental changes in rates of depression.

We suggest two high-priority areas of study to advance our understanding of the cognitive vulnerability contagion phenomenon and, in turn, risk for depression. One priority is to determine the scope of the contagion effect. This study used a unique sample and study design to maximize sensitivity for detecting the hypothesized contagion effect. Participants in our study were experiencing a dramatic change in their social context. For many of them, moving away to college was the first time that they had to live apart from their family and friends for an extended period. In addition, participants were randomly assigned to roommates. This naturalistic experimental enabled us to make definitive conclusions about the presence of a contagion effect and rule out alternative explanations such as a self-selection bias, in which participants choose their own social context. The sample and experimental design are significant strengths of the study as they optimized our ability to detect the contagion effect and make strong conclusions about our results. At the same time, however, this methodology does not capture the typical person’s social environment. Thus, research is needed to determine the generalizability of the contagion effect. One strategy is to test whether the findings apply to other life transitions such as moving to a new city or entering a nursing home. It is likely that the contagion effect will generalize to such situations because they are akin to moving to college. They represent a new life stage (e.g., starting a first “real job”) and a major transition to a new living environment.

However, given the relatively short amount of time in which cognitive vulnerability changed in the present study, we suspect that the contagion effect may not be restricted to major life transitions. Indeed, we posit that a contagion effect may be detected even in a much more ubiquitous social context—a person’s social media network. Social media platforms such as Facebook and Twitter are changing the interpersonal landscape. One in seven people are active Facebook users (Facebook Inc., 2012), and more than 100 million people use Twitter (Miller, 2011). People are now able to develop relationships (both platonic and romantic) with millions of individuals from all over the world. This means that a person’s social milieu has the potential to be much larger and much more fluid than ever before. No longer is a person’s social environment restricted to his or her immediate surroundings. It is now possible to alter one social network with the click of a button (e.g., adding a new “follower” on Twitter or accepting a friend request on Facebook). We suspect that as a person’s social media network changes over time, so will his or her levels of cognitive vulnerability. Preliminary research has already demonstrated the potential power of social networks in creating contagion effects. For example, scientists have found social media networks to exhibit contagion-like effects for mood, smoking habits, and even obesity (Christakis & Fowler, 2013). A recent study of reciprocal-reply networks on Twitter found that happiness appears to be contagious up to three degrees away from the user (Bliss, Klourmann, Harris, Danforth, & Dodds, 2012). These results add to a growing number of studies that suggest that psychological constructs can be contagious within social media networks and that the effects are not merely due to homophily (Christakis & Fowler, 2013).

Social media networks may be particularly influential in altering cognitive vulnerability levels due to their increased activity during major life events. For example, deaths of high-profile figures, natural disasters, and other highly stressful events are among the most tweeted events of each year. These types of events are fodder for cognitive vulnerability. During these times, social media users are exposed to the cognitive reactions and coping strategies of potentially hundreds of people around the world. A question that is ripe for future research is whether the tone and consistency of comments in individuals’ social media network have the potential to alter their cognitive vulnerability levels over their life span and, in turn, influence their risk for depression.

A second priority for future research is to determine if the contagion effect can be leveraged to create resilience to depression. Until now, it appeared that the only way to alter cognitive vulnerability to depression was to target it directly via cognitive therapy. However, our findings suggest that it may be possible to use an individual’s social environment as part of the intervention process (either as a supplement to existing cognitive interventions or possibly as a stand-alone intervention). An intervention that capitalizes on the contagion effect would require an evaluation of the patient’s interpersonal context including family, friends, and perhaps even his or her social media network. The primary purpose of this evaluation would be to ensure that the patient is surrounding himself or herself with an adaptive cognitive social context. The therapist could assess if people in the patient’s life are modeling and providing adaptive cognitive feedback about stress and negative life events. The therapist could then provide those with maladaptive cognitive styles with information about the contagion effect along with training that would help them identify maladaptive thought patterns and provide examples of more adaptive ways of thinking (see Dobkin et al., 2007, for an example of a program designed to train partners in adaptive cognitive feedback). Surrounding a person with others who exhibit an adaptive cognitive style should help to facilitate cognitive change in therapy.

A second benefit of developing interventions that capitalize on the contagion effect is that such interventions could be used in experimental designs to determine the mechanisms by which contagion occurs. There are a number of social processes that might contribute to the contagion of cognitive vulnerability. For example, people might contract high levels of cognitive vulnerability by observing (or modeling) others around them. Another possibility is that a person learns maladaptive cognitive styles on the basis of inferences that significant others make for his or her (i.e., the person’s) behavior. In this case, the person develops a cognitive style consistent with the feedback that he or she receives. By using an experimental intervention design, it will be possible to determine which of these processes underlie the contagion effect. Participants could be randomly assigned to different social context influences such as direct feedback, modeling, or positive self-disclosure. Identifying more precisely the mechanisms driving the contagion effect would enable researchers to create a more focused, potent, and time efficient intervention.

Although the goal of the present study was to test the hypothesis that cognitive vulnerability could be contagious, the findings also call into question the literature on the contagion of depressive symptoms. Prior research suggests that depression itself might be contagious between individuals (Joiner & Katz, 1999). However, we did not detect a contagion effect for depressive symptoms. Roommates’ baseline levels of depressive symptoms were not significantly associated with the participants’ levels of depressive symptoms at either 3 months or 6 months. Rather, the strongest predictors of prospective changes in participants’ level of depressive symptoms were (a) their own baseline level of depressive symptoms, (b) their own baseline level of cognitive vulnerability, and (c) their roommates’ baseline level of cognitive vulnerability. Thus, the results of this study suggest that the contagion is not depression per se, but rather the cognitive factors that increase risk for depression. Although it is true that contracting higher levels of cognitive vulnerability might then lead individuals to develop more depression, this increase in depression would not be due to their roommate’s level of depression, but rather to their roommate’s cognitive vulnerability. In addition to this indirect effect via cognitive vulnerability, there are also at least two additional differences between the present study and previous studies that may explain why we did not find a depressive contagion effect. First, we used a more appropriate data analytic strategy for testing dyadic effects (APIM; Kenny, 1996). Prior studies testing depressive contagion effects have relied on traditional statistical techniques, which assume the nonindependence of observations. Because the nonindependence assumption is often violated in studies of close interpersonal relationships, the test statistics in previous studies may have been biased (Cook & Kenny, 2005). Second, we had random assignment of participants to roommates. Prior studies testing depressive contagion effects did not randomly assign participants to dyads, so these studies cannot rule out the influence of participants self-selecting their interpersonal environments in their effects.

It is important to note that the rumination vulnerability factor was less stable over time (and more likely to be contagious) than the hopelessness theory vulnerability factor. The rumination vulnerability factor was contagious at both the 3-month and 6-month time points, whereas the hopelessness theory vulnerability was not. These results are consistent with a study by Hankin (2008), which showed that the hopelessness theory cognitive vulnerability factor was highly stable over time, whereas the rumination vulnerability factor exhibited more moderate stability. One explanation for why the two vulnerabilities exhibit different temporal stabilities is that they focus on different cognitive attributes. The hopelessness theory vulnerability factor focuses on cognitive content (i.e., the particular types of interpretations that people generate about stress), whereas response styles theory focuses on the process by which people respond to negative moods (i.e., brooding vs. distraction). We suspect that cognitive processes might be more amenable to change than cognitive content. For example, learning a new process such as distraction (e.g., playing a sport rather than brooding) in response to a negative mood seems more readily attainable for someone to learn than altering his or her cognitive content, which is likely part of a well-organized latent schematic structure. However, the issue of stability remains an empirical question. Future research is needed to determine why the hopelessness vulnerability factor is more stable than the rumination vulnerability.

In conclusion, the goal of this study was to establish the presence of the contagion phenomenon for cognitive vulnerability to depression. Although it is difficult to fathom the idea of people suddenly “catching” a different style of interpreting their world, the current study demonstrates that people’s cognitive vulnerability can affect the cognitive vulnerability of those around them, at least when people are going through relatively large-scale changes in their interpersonal environment. Using a natural experimental design with three time points, we found that when people change from a home context to a college context they are susceptible to the influence of the cognitive styles of those around them, even when the others are initially strangers. This effect was shown at both 3 and 6 months for the RRS, and the results cannot be explained by a depression contagion effect or by increased stress. Moreover, the contagion effect has depressive consequences. Those who experienced an increase in cognitive vulnerability had increased risk for future depressive symptoms. These results call for a change in how scientists conceptualize cognitive vulnerability and provide clues for understanding developmental fluctuations in risk and resilience to depression.