Desperately Seeking Support

Abstract

Much is unknown about adult attachment style formation. We investigate whether negative reinforcement schedules promote hallmark features of secure and anxious attachment styles in a shock threat support-seeking paradigm. Participants ostensibly asked for help from another participant seated in another room. Each time a shock threat signal appeared they were to press a button to indicate their need for help. The supporter could then stop the imminent shock. The reliability of the supporters was varied such that some supporters were consistent (continuous reinforcement) whereas others were inconsistent (variable ratio reinforcement). Results indicated that inconsistently responsive others, reinforcing on a variable ratio schedule, led to heightened approach-related attentional biases toward the supporter, measured by event-related potentials, increased positive attachment associations with the supporter, implicitly measured via a lexical decision task, and more negative explicit evaluations of the supporter.

Keywords

attachment emotion social cognition interpersonal processes person perception

The need for social connection permeates human life (Baumeister & Leary, 1995; Beckes & Coan, 2011; Bowlby, 1969/1982; Holt-Lunstad, Smith, & Layton, 2010). John Bowlby’s attachment theory may be the most extensive and durable theory of human connection. Originally focused on child–parent relationships, attachment theory has been subsequently extended to adult relationships (Hazan & Shaver, 1987). Adult attachments come in many types and most theorists argue for two dimensions of attachment quality, anxiety, and avoidance (Mikulincer, Shaver, & Pereg, 2003). Typically, scholars classify individuals into three categories of attachment: secure, anxious, or avoidant (cf. Fraley, Waller, & Brennan, 2000). Secure individuals comfortably seek support and have optimistic views of relationships and social support. Anxious individuals tend to express fear of abandonment, ambivalence about support, clingy behavior, and pessimism about receiving support. Avoidant individuals, alternatively, tend to discount their need for support and avoid dependence on others.

Adult attachment styles are linked to earlier attachment dynamics from infancy through adolescence (e.g., Simpson, Collins, Tran, & Hayden, 2007) and have stable elements over time (Fraley, Vicary, Brumbaugh, & Roisman, 2011) but also vary as a function of specific relationships (Klohnen, Weller, Luo, & Chloe, 2005). Whereas the stable elements predict many outcomes, including the development of relationship-specific attachment styles, several relationship outcomes are primarily linked to relationship-specific attachment styles, which can differ from dispositional styles. Yet, little is known about how relationship-specific attachment styles develop. Here, we test whether the psychological features of different styles can be promoted in novel relationships through conditioning by manipulating negative reinforcement schedules (Beckes & Coan, 2015). If distinct negative reinforcement schedules lead to response patterns related to disparate attachment styles, then relationship-specific attachment styles may emerge from these basic learning processes through repeated iteration. Examining these processes could provide a foundation for linking mechanisms to descriptive data and provide some experimental data to compare to extant models of attachment formation (Beckes & Coan, 2015; Mikulincer et al., 2003).

The Origins of Individual Differences in Attachment

Mikulincer and Shaver (2005) argue for a support-seeking dynamic in which individuals learn to use different strategies to manage distress, which normatively promotes proximity seeking to attachment figures. If the attachment figure is responsive to that distress, then the individual experiences well-being and safety (secure pattern). If the attachment figure is inalterably unresponsive, then the individual will employ deactivating strategies, distancing oneself from threat and attachment-related cues (avoidant pattern). If the attachment figure is unresponsive, but can be coaxed into providing support, then the individual will employ hyperactivating strategies such as clinginess, chronic attachment system activation, and experience ambivalent emotional responses (anxious pattern). When repeated over time, the individual will develop a style within the relationship.

Beckes and colleagues (Beckes & Coan, 2013, 2014, 2015; Beckes, Coan, & Morris, 2013; Beckes, Ijzerman, & Tops, 2015; Beckes, Simpson, & Erickson, 2010) argue that the formation of attachment styles can be understood as a negative reinforcement learning process with social responsiveness acting as an unconditioned reward (cf. Harlow, 1958). From this perspective, continuous negative reinforcement (consistent responsiveness) should produce secure patterns. Such consistency produces predictable contingencies between behavior and outcomes. As a result, support seeking occurs when needed, but not when unneeded, evaluations of consistent supporters become unambiguously positive and approach motivation toward that provider is enhanced.

Alternatively, variable ratio schedules (inconsistent responsiveness) should produce ambivalent attachment attitudes and hyperactivation, such as chronic support seeking and strong approach motivation. This matches with long established findings that variable ratio reinforcement produces steep response slopes in animal learning (Skinner, 1956) and increases the incentive salience of rewards (Anselme, Robinson, & Berridge, 2013). For example, unpredictable food rewards increase rats’ approach speed, effort to achieve the reward, and vigor of engagement with conditioned cues.

Conditioning and Attachment Attitudes

In line with this approach, Beckes, Simpson, and Erickson (2010) and Beckes, Coan, and Morris (2013) found that repeatedly presenting a threat followed by a warm, smiling face led the face to prime the recognition of secure attachment-related words and to inhibit the recognition of insecure attachment-related words. This suggests that continuous negative reinforcement promotes positive implicit associations. Similarly, inconsistent support should be associated with relief, calm, and so on, just unpredictably and that unpredictability should increase the incentive salience (Anselme et al., 2013) of that support, which should increase the strength of the association between the inconsistent person and implicit security constructs. But what about implicit insecurity constructs?

Dispositionally, insecure individuals tend to have attachment representations of both a negative and positive valence chronically accessible (e.g., Mikulincer, Gillath, & Shaver, 2002; Zayas & Shoda, 2005, 2014). One possibility is that variable ratio conditioning immediately produces associations like those seen in dispositional attachment anxiety, the dispositional hypothesis. But will those dispositional findings translate to the very first moments of learning within a relationship? It is plausible that conditioning primes positive attachment constructs and inhibits negative attachment constructs primarily through attachment motivation being satisfied via reinforcement and does not immediately produce accessibility to bivalent attachment constructs. Thus, the associations may be a function of attachment motivation and reinforcement, not the manifestation of dispositional attitudes (the motivation and reinforcement hypothesis).

Either way, inconsistent responsiveness should promote explicit evaluations of supporters as untrustworthy and unpredictable, whereas consistent responsiveness should be viewed as trustworthy and predictable. Thus, inconsistency should produce ambivalence whether the dispositional or motivation and reinforcement predictions are correct. If the motivation and reinforcement hypothesis is correct, it seems likely the tension between intense approach motivation and explicit distrust ultimately promotes dispositional bivalent attachment associations seen in anxious attachment.

Conditioning and Approach Motivation

Evidence for increased approach motivation toward supporters after negative reinforcement comes from electroencephalography (EEG) and event-related potential (ERP) studies. In particular, the P1 potential, a positive deflection peaking around 100 ms after face presentations, is associated with emotional attention biases. This component emerges early in processing and is associated primarily with attention modulation via external stimuli. Cunningham and colleagues (Cunningham, Van Bavel, Arbuckle, Packer, & Waggoner, 2012) suggest that social approach motivation modifies P1 responses. P1s are greater to faces toward which participants have an intrinsic approach bias, and increasing approach, as opposed to avoidance, motivation increases P1s. As such, the P1 may be useful in investigating whether negative reinforcement in attachment contexts increases P1 responses to conditioned faces given that the conditioning procedure is designed to promote the approach behavior of support seeking. Supportive of these predictions, Beckes et al. (2013) found increased P1 responses to continuously negatively reinforced faces indicating increases in approach-related attentional biases. But what about variable ratio schedules? P1s should be greatest to unpredictable supporters given that unpredictable reinforcement increases approach motivation.

Experimental Design and Predictions

We employed a shock threat conditioning paradigm, during which participants asked for help from a “supporter” to prevent being shocked. We manipulated, within-subjects, the reinforcement schedule such that some supporters were consistently helpful (continuous reinforcement), inconsistently helpful (variable ratio reinforcement), and novel at test (neutral). We had three primary dependent variables (DVs), implicit attachment attitudes toward each face (response times [RTs] in a lexical decision task to secure and insecure attachment words), explicit attitudes toward each face, and approach-related attentional biases (P1 ERP component) to each face. We chose these DVs because they allowed us to test for attitudinal ambivalence and approach-related attention.

During test we repeatedly presented each face as a prime in a lexical decision task. To measure both positive and negative implicit attachment associations, we employed secure attachment words and insecure attachment words. In addition, we measured nonattachment words to ascertain whether learning was restricted to the attachment domain. The lexical decision task was a 3 conditioning schedule (continuous vs. variable ratio vs. neutral) × 2 word type (attachment vs. non-attachment words) × 2 valence (positive vs. negative words) within-subjects design. For the P1 responses, we tested left and right hemisphere electrodes resulting in a 3 conditioning × 2 electrode hemisphere within-subjects design. Finally, we measured explicit ratings of each face, leading to a three-level conditioning within-subjects design.

For each DV, we had a priori hypotheses (see Open Science Framework [OSF] website for registered predictions, materials, design, and code: https://osf.io/erk9c, preregistration materials were consolidated before any data were processed, but shortly after data collection had started, see Table 1 for predictions and outcomes). For RTs, we predicted a three-way interaction between conditioning, word type, and valence, driven by an interaction between conditioning and valence in the attachment words, with no effects on the nonattachment words. We then produced paired-comparison predictions. For the variable ratio condition, we predicted primes would facilitate recognition of secure words: (1) variable ratio primed secure word RTs < continuous primed secure word RTs and (2) variable ratio primed secure word RTs < neutral primed secure word RTs. For insecure words, the dispositional hypothesis would predict (1) variable ratio primed insecure word RTs < continuous primed insecure word RTs and (2) variable ratio primed insecure word RTs < neutral primed insecure word RTs, whereas the motivation and reinforcement hypothesis would predict (1) variable ratio primed insecure word RTs > continuous primed insecure word RTs and (2) variable ratio primed insecure word RTs > neutral primed insecure word RTs. For the continuous conditioning, we predicted (see Beckes et al., 2010, 2013) (1) continuous primed secure word RTs < neutral primed secure word RTs and (2) continuous primed insecure word RTs > neutral primed insecure word RTs.

Table 1.
Predictions at Each Level of Analysis for Each Dependent Variable, Fit With Data by Standard Statistical Cutoffs, Fit With Data by General Pattern, Number of Tails for the Test (Applicable Only to Paired t-tests), and p Value of the Resultant Test.

Test Type Specific Prediction Fit Prediction by p Cutoff Fit Prediction by Pattern Tails^a p Value

Response times

Omnibus three-way interaction Conditioning × Word Type × Valence interaction Yes Yes N/A .031

Nested two-way interactions

Attachment words Conditioning × Valence interaction Yes Yes N/A .028

Nonattachment words No Conditioning × Valence interaction (null) Yes Yes N/A .506

Nested paired-comparisons

Positive attachment words Variable ratio < continuous No Yes 1 .06

Variable ratio < neutral Yes Yes 1 .0025

Continuous < neutral No Yes 1 .108

Negative attachment words Variable ratio ≠continuous No N/A 2 .48

Variable ratio ≠ neutral No N/A 2 .417

Continuous > neutral No Yes 1 .464

Explicit evaluations

Omnibus ANOVA Conditioning main effect Yes Yes N/A <.001

Nested paired-comparisons Variable ratio < continuous Yes Yes 1 <.001

Variable ratio < neutral Yes Yes 1 <.001

Continuous > neutral Yes Yes 1 <.001

P1 ERPs

Omnibus ANOVA Conditioning main effect Yes Yes N/A .041

Conditioning × Hemisphere interaction No Yes N/A .607

Nested paired-comparisons Variable ratio > continuous Yes Yes 1 .035

Variable ratio > neutral Yes Yes 1 .0085

Continuous > neutral No Yes 1 .237

Note. Given the a prior nature of the hypotheses, and their empirical and theoretical basis, we believe that this approach was the most principled and appropriate. ANOVA = analysis of variance; ERPs = event-related potentials.

^aOne-tailed tests were used for specific directional hypotheses as is recommended by statistical experts (e.g., Cho & Abe, 2013).

For the P1 ERP, we predicted a significant Conditioning × Hemisphere interaction and a main effect of conditioning on P1s. Specifically, variable ratio conditioned faces would produce the largest P1 responses, continuously conditioned faces the second largest P1 responses, and neutral faces the smallest P1 responses. Further, we expected this pattern to be strongest in the left hemisphere as found in Beckes et al. (2013).

Finally, it was predicted that variable ratio faces would be explicitly evaluated more negatively than neutral faces due to their unpredictability and unreliability, and continuously reinforced faces would be evaluated more positively than neutral faces.

The pattern of results we predict from the variable ratio schedule would show the type of motivational hyperactivation and ambivalence one might expect within the early development of a new relationship-specific style when the new partner is inconsistently responsive. Alternatively, the type of moderate approach motivation and unambiguous positive attitudes predicted to emerge out of the continuous schedule matches what we would expect of an early developing secure relationship.

Method

Participants

Sixty participants (41 women), 18–43 years old (M = 19.5), participated. Sixty-five percentage identified as White, 8% African American, 10% Asian, 12% Hispanic, and 3% Other. Two participants were left-hand dominant, and four spoke English as a nonprimary language. Participants were compensated (extra credit or cash) for participation and treated in accordance with all institutional review board requirements. We invoked a stop rule at 60 participants with complete data. This rule was based on a sample size estimation using GPower paired t-test calculator (all manipulations were within subjects) assuming a modest effect size of dz = .4 and power of .80, needing around 60 participants.¹

Procedure

Setup

After informed consent, we delivered a cover story that participants were in a study of the provision and receipt of social support with three other participants. We told them that they would at times act as a supporter of another participant and would be able to prevent the other person from getting an electric shock and at other times would be supported by another participant who could prevent them from getting a shock. We took a digital photo of the participant (immediately erased) to prevent participant suspicion upon seeing photos of “other participants” in a later task. Subsequently, they completed demographics questions, the Experience in Close Relationships Revised (ECRR, attachment style measure), and researchers placed EEG electrodes on them. They then followed an Eprime practice script with instructions on the main tasks.

EEG preparation

EEG recordings included 30 scalp electrodes measured via the BrainVision ActiChamp (Version 32; BrainVision LLC, 2016) battery-powered system. We used a headcap with electrode sites in extended international 10/20 locations (Fp1, Fp2, F3, F7, Fz, F4, F8, FC5, FC1, FC2, FC6, T7, C3, Cz, C4, T8, CP5, CP1, CP2, CP6, P7, P3, Pz, P4, P8 O1, OZ, O2, TP9, TP10). Fpz served as the ground, and Cz as a reference. SuperVisc 1000 g electrode gel reduced impedance between the scalp and Ag–AgCl electrodes.

Supportive task

After setup, participants experienced the supportive task. We designed this task to increase the believability of subsequent procedures by having the participant acts as a supporter before being supported. A math sheet was given for the participant to complete, and they were told to watch for signals over the computer (a red box) while completing the math problems. The math task was designed to provide a plausible reason that someone might miss a signal for help from another participant. If they saw a signal, they were told they could prevent a shock from being delivered to the other participant by pressing a button–box button. Details are in the Supplemental Materials.

Learning task

During learning, participants saw red X’s, indicating imminent shock if the supporter failed to help, and blue O’s, indicating safety. A red X meant they should push a button-box button as frequently as needed to signal to a supporter in another room. This signaled the supporter to push a button to prevent the shock (see Figure 1). This task had two blocks. In one block, the supporter was consistently responsive (continuous reinforcement). In the other block, a different supporter was inconsistently responsive (variable ratio reinforcement). Block order and the block-supporter image pairings were completely counterbalanced across participants. Details of the instructions, stimulus presentation times, and supporter response profiles are available in the Supplemental Materials.

Figure 1.
Depiction of the sequence of events during the learning task for both the inconsistent variable ratio learning condition and the consistent continuous reinforcement condition. Each trial begins with the presentation of an “X” indicating threat or “O” indicating safety. Participants are then allowed to push a button on a button box to ostensibly ask for help from another participant. In the variable ratio condition, the other responds with varying speed in 75% of trials but fails to respond in 25% resulting in a shock. In the continuous condition, the other responds quickly and in 100% of trials.

Lexical decision task

Test trials followed learning. Researchers recorded EEG data and the participant started the lexical decision task. Upon receiving and understanding the instructions (see Supplemental Materials), participants did three practice trials and began. Stimulus presentations had a black background with white text. Trials included a 500 ms fixation point, followed by a 1000 ms prime face, and then by a word or nonword until a word/nonword judgment was made. Participants pushed the first button on the button-box for words and second button for nonwords as quickly as possible. Each word/nonword was paired with each face once and pairings randomly presented.

Final tasks

Subsequently, participants completed Self-Assessment Manikins (SAM) Scales measuring affective arousal and valence during shock and explicit ratings of each face (face-item pairings randomized). We then asked the participant to speculate on the predictions of the study, if anything appeared odd, and if anything seemed suspicious (suspicion was noted in the log notes). Finally, the participant was debriefed and thanked.

Materials

SAM Scales

SAM are a visual self-assessment self-report measure of valence and arousal (Bradley & Lang, 1994). Scales depict five anthropomorphic manikins as a visual Likert-type scale (e.g., valence includes a broadly smiling figure for very positive and a deeply frowning figure on the for very negative). SAM scales were used as a manipulation check to ensure effectiveness of the threat.

ECRR

The ECRR is a self-report measure of attachment style originally developed by Brennan, Clark, and Shaver (1998) and revised by Fraley, Waller, and Brennan (2000). It includes 18 anxiety items (e.g., “I often worry that my partner will not want to stay with me”) and 18 avoidance items (e.g., “I find it difficult to allow myself to depend on romantic partners”), each rated on a 7-point Likert-type scale. The ECRR exhibited high internal consistency reliability (avoidance α = .90; anxiety α = .92).

Explicit Rating Scale

Participants rated faces on 8 positively valenced adjective items: trustworthy, predictable, kind, responsive, thoughtful, reliable, likable, and attractive. The rating scale involved agreeing or disagreeing with the degree to which each face possessed each quality on a 1–5 scale (1 = strongly disagree, 5 = strongly agree). Internal consistency reliability was high for faces in each conditioning group (variable α = .91; continuous α = .93; neutral α = .87).

Face stimuli

Face stimuli were taken from the Facility of Industrial Engineering (FEI) face database (http://fei.edu.br/∼cet/facedatabase.html). Three smiling female faces were used in this study (15-12.jpg, 79-12.jpg, and 99-12.jpg). Practice items included an unrelated male face.

Lexical decision word lists

The lexical decision task consisted of the presentation of positive attachment words (e.g., secure, nurture), negative attachment words (e.g., avoid, cold), positive nonattachment words (e.g., lust, intercourse), negative nonattachment words (e.g., disease, rot), neutral words (e.g., gallon, tissue), and nonwords (e.g., bunturn, joov). These words were taken from previous research by Beckes et al. (2010, 2013). Readers can refer to the OSF website for all word lists https://osf.io/erk9c.

Math distracter

The math distractor consisted of 12 problems (e.g., If f(x) = 3x ³ − 7x + 5, then f(−1) =). This task was designed to be too difficult (some unsolvable) to complete in the time given.

Analysis and Data Reduction

RT data

RTs were averaged per participant in each of the lexical decision conditions after removing outliers (below 300 ms and 3 SD above the mean, consistent with Beckes et al., 2013) and trials with incorrect responses (average accuracy was high at 92%). Data were analyzed in SPSS (Version 22; IBM Corp., 2013) using repeated measures analysis of variance (rmANOVA) and further interrogated with rmANOVA and paired t-tests.

ERP analysis and data reduction

EEG was sampled at 500 Hz, processed in electroencephalography lab (EEGLAB; Delorme & Makeig, 2004) and event related potential lab (ERPLAB; Markley, Luck, & Lopez-Calderon, 2011). Data were filtered offline with an infinite impulse response (IIR) Butterworth, 1 Hz high-pass filter and a 30 Hz low-pass filter. An average reference was derived offline. Electrodes placed below the left and the right of the right eye that captured eye movements and blink artifacts. Epochs were defined as 200-ms prestimulus presentation to 800-ms poststimulus presentation. ERPLAB’s moving window peak-to-peak artifact detection function detected artifacts (voltage threshold of 80 μv). Visual inspection verified this cutoff. Data were corrected to the mean voltage of the prestimulus interval. Eight participants failed to have at least 20 accepted trials for each face stimulus and were removed from the analysis. Mean P1 amplitudes were extracted in each target electrode between 110 and 140 ms after stimulus onset. Decisions on filters, average referencing, artifact detection, and the use the average amplitude between 110- and 140-ms poststimulus presentation were identical to those used in Beckes et al. (2013). Specific analysis plans are documented on the OSF page https://osf.io/erk9c in the analyses and predictions document.

Results

Manipulation Check SAM Scales

One sample t-tests were performed on the SAM Scale data. We tested the mean of the sample against values that would represent no arousal (5) and neutral valence (3). Arousal was significantly increased by shock (5 = no arousal, 1 = highest arousal), t(59) = −14.00, p < .001, 95% confidence interval (CI) [−2.61, −1.96], (M = 2.72, SD = 1.26), indicating moderate to high arousal. Valence had a test value of 3, which indicates neutral valence* (5 = most positive, 1 = most negative). Mean valence was significantly lower than 3, t(59) = −6.42, p < .001, 95% CI [−0.94, −0.49], (M = 2.28, SD = .87), indicating a negative response to the shock.

Lexical Decision Task

The lexical decision task involved priming words of different categories (word type) and valence with conditioned faces. We predicted a three-way interaction between conditioning of the priming face (variable ratio vs. continuous vs. neutral), word type (attachment vs. nonattachment), and word valence (positive vs. negative). Significant results emerged for the main effect of word type, F(1, 59) = 33.32, p < .001, η_p ² = .361, with faster RTs to attachment words (M = 601.45, SE = 1.63)² than nonattachment words (M = 620.26, SE = 1.64), 95% CI [−25.33, −12.29]. Finally, the predicted interaction between conditioning by word type by valence was significant, F(1.98, 116.96) = 3.61, p = .031, η_p ² = .058 (for descriptive statistics, see Figure 2 and Table 2). No main effect of conditioning was detected, F(1.80, 106.13) = .69, p = .489, η_p ² = .012, no main effect of valence was detected, F(1, 59) = 3.83, p = .055, η_p ² = .061, nor were any other interactions: Conditioning × Word Type, F(1.88, 110.96) = .70, p = .492, η_p ² = .012; Conditioning × Valence, F(1.97, 116.33) = .26, p = .769, η_p ² = .004; Word type × Conditioning, F(1, 59) = .40, p = .531, η_p ² = .007.

Figure 2.
Mean and within-subjects 95% confidence interval of response times in milliseconds to words primed by faces within each condition of the design. Face primes included the faces negatively reinforced by variable ratio or continuous primes in the learning trial and a novel neutral face. Types of words are split up by valence and word type (attachment-related positive:secure words, attachment-related negative:insecure words, nonattachment positive:sexual desire words, and nonattachment negative:disgust words).

Table 2.
Means and Within-Subjects Standard Errors or Lexical Decision Response Times by Condition.

Schedule Attachment Words Nonattachment Words

Positive Negative Positive Negative

M SE M SE M SE M SE

Variable ratio 587.78 3.69 609.36 5.25 619.14 6.02 616.03 6.47

Continuous 597.84 4.69 604.32 3.80 620.76 6.20 620.45 6.15

Neutral 605.63 5.28 603.78 4.22 613.92 4.28 623.26 6.81

We further interrogated these effects to make paired-comparisons and test nested models. For word space reasons, these subanalyses are in the Supplemental Materials. These analyses indicated that the interaction between conditioning and valence was significant for the attachment words, but not the nonattachment words, comporting with predictions. Further analyses indicated a pattern of RTs to secure words consistent with predictions, but only the variable ratio versus neutral comparison met threshold. We found no significant effect for the insecure words (for a summary of these results, see Table 1).

Explicit Evaluation

For explicit evaluations, we predicted a main effect of conditioning such that the variable ratio condition would result in the most negative evaluations and the continuous condition would result in the most positive evaluations. A three-factor within-in subjects ANOVA revealed a significant difference in explicit evaluations as a function of conditioning, F(1.56, 92.18) = 33.97, p < .001, η_p ² = .365, see Figure 3. As predicted, the continuous reinforcement faces (M = 4.05, SE = .09) were rated most positively, followed by the neutral faces (M = 3.45, SE = .05) and finally the variable ratio faces (M = 2.95, SE = .09). All pairwise comparisons were significant: variable ratio versus continuous, t(59) = −6.66, p < .001, 90% CI [−1.38, −0.82]; variable ratio versus neutral, t(59) = −4.47, p < .001, 90% CI [−0.69, −0.32]; continuous versus neutral, t(59) = 5.12, p < .001, 90% CI [0.40, 0.79].

Figure 3.
Mean and within-subjects 95% confidence interval of mean explicit evaluations on a 5-point Likert-type scale within each condition of the design (variable ratio, continuous, and neutral).

ERP P1

Electrodes O1, O2, P3, and P4 were selected for analysis based on previous research (Beckes et al., 2013). We conducted rmANOVA with conditioning (variable ratio vs. continuous vs. neutral) and hemisphere (left vs. right) as the independent variables (IV). We predicted a main effect of conditioning such that variable ratio face P1s would be greater than P1s to continuous faces, which would be greater than P1s to the neutral faces. Moreover, we predicted an interaction between conditioning and hemisphere with an enhanced main effect in left-hemisphere electrodes. A significant conditioning main effect emerged, F(1.98, 98.94) = 3.31, p = .041, η_p ² = .062. The pattern of results conformed to predictions, and two pairwise comparisons reached statistical significance. The variable ratio conditioned faces (M = 5.63, SE = .10) led to significantly greater activity than the continuously reinforced faces (M = 5.31, SE = .11), p = .07, p = .035 one-tailed, 90% CI [0.031, 0.618]. The variable ratio conditioned faces led to significantly greater activity than the neutral faces (M = 5.17, SE = .11; p = .017, p = .0085 one-tailed, 90% CI [0.149, 0.776]). The continuously conditioned faces were nonsignificantly greater in activity than the neutral faces (p = .473, p = .237 one-tailed, 90% CI [−0.182, 0.459]; see Figure 4). The predicted interaction between conditioning and hemisphere did not emerge, F(1.82, 91.02) = .48, p = .605, η_p ² = .009.

Figure 4.
Depicts the mean and 95% confidence intervals of the P1 amplitude in the 110–140 ms range (A) for electrodes P3, P4, O1, and O2 to faces that were either novel (neutral) or negatively reinforced with a variable ratio or continuous schedule and average event-related potential response to faces in the O1 electrode (B) by condition.

Exploratory Analyses

Additional exploratory analyses were conducted but only briefly mentioned here. We explored the relationship between the DVs as a function of condition. Only one significant effect emerged for the correlation between the continuous condition P1 difference and the continuous condition secure word speed, r(51) = .321, p = .022. In addition, given that general attachment styles are likely predictors in the development of relationship-specific attachment styles, we found evidence of their influence through the ECRR. Moreover, we explored the impact of gender on our conditioning data. Additionally, to ensure that suspicious participants were not driving any of the effects found, several analyses were performed with suspicion included as an independent variable. Suspicion was not a significant predictor in the lexical decision or ERP analyses (ps > .05). Suspicion did significantly interact with conditioning on the explicit ratings, F(1.55, 90.05) = 5.076, p = .014, η_p ² = .080, but the effects held whether the participants were suspicious, F(1.351, 24.31) = 20.27, p < .001, η_p ² = .530, or not, F(1.65, 65.98) = 17.80, p < .001, η_p ² = .308. All analyses are available in the Supplemental Materials.

Discussion and Conclusion

This experiment sheds light on the initial stages of relationship-specific attachment formation exploring the hypothesis that negative reinforcement schedules are an important ingredient in the development of relationship-specific adult attachment styles. It was hypothesized that variable ratio negative reinforcement schedules should produce attitudinal ambivalence and hyperactivated approach motivation toward a supporter, whereas continuous reinforcement schedules should produce unambiguous positive attitudes and moderate approach motivation toward a supporter. Here, variable ratio schedules produced features of attachment anxiety as predicted, including intensified approach motivation and attitudinal ambivalence toward a novel supporter. All a priori predictions of paired-comparisons between the variable ratio condition and neutral condition were confirmed, except for our prediction regarding the insecure word RTs. This may be due to sampling variation, something about the particular methodological details of this study or could be due to a true null difference. Moreover, we also detected greater P1 responses to the variable ratio faces than to the continuous faces, and less positive evaluations of the variable ratio faces relative to the other two conditions. The P1 results indicate an increase in approach motivation as a function of variable ratio conditioning, whereas the facilitation of secure word RTs and negative explicit evaluations indicates attitudinal ambivalence.

Although the variable ratio findings are generally supportive of our model, caution should be taken in interpreting these results. First, other measures of approach motivation should be used to verify our interpretation of the P1 effects found here, as only converging operations and findings can provide evidence for a truly strong interpretation. Second, these effects are constrained to specific relationship attachment strategies in the earliest stages of development, and a very particular type of interaction. The type of support being provided here is primarily instrumental, but attachment processes often play out in more emotionally supportive contexts, such as giving someone a hug after a bad day. As such, these findings must be supplemented with more ecologically valid data before generalizing beyond this context. Alternatively, the data are high in internal validity and allows causal inferences that may be difficult in naturalistic studies. Ideally, future research will bridge this gap with ecologically valid experiments with enough control to have moderate external and internal validity.

The data are also consistent with the hypothesis that continuous schedules produce secure associations, with P1 responses and secure concept priming intermediate between controls and variable ratio conditioned faces, and more positive explicit evaluations than controls, but outside of the explicit ratings the effects were unconvincing and failed to reject the null hypothesis. To examine the degree to which the continuous conditioning data comport with past studies, we included a Bayesian analysis in the Supplemental Materials that examines the RT data in conjunction with data from Beckes et al. (2010, 2013; see Supplemental Tables 5 and 6 and Supplemental Figure 1). This indicates the current data modestly strengthens the evidence for an effect of continuous negative reinforcement on implicit attachment attitudes (BF₁₀ = 8.7 up from BF₁₀ = 6.0), and the BF₁₀ indicates substantial evidence for the alternative versus null hypothesis (Jeffereys, 1961). These results indicate a possible lack of power, likely due to two factors. First, sample size was restricted due to EEG costs. Future studies with less time and cost-intensive measures should collect larger sample sizes. Second, some participants may not have perceived a real threat in the continuous condition given the predictability of the consistent responder. Future studies could include an occasional shock that the partner cannot stop in order to maintain shock credibility. Future studies should also consider the inclusion of other measures of approach motivation, and the development of more precise standardized measurements.

This data should not be interpreted as evidence that global attachment styles lack stability, or that this method is automatically capable of modifying either global or already intact relationship-specific attachment styles. Future research should attempt to find ways to explore these possibilities. Exploratory analyses of general attachment anxiety and avoidance (see the Supplemental Material for details) replicated several previous findings in the literature, such as anxious individuals responding generally faster to bivalent attachment constructs, avoidant individuals slower to respond to all constructs, and avoidant attachment altering P1 responses to emotional faces (Dan & Raz, 2012). These results indicate the possibility that anxious individuals respond to social stimuli with vigilance increasing RTs, whereas avoidant individuals respond with either disfluency in processing social information, or an attempt to suppress automatic responses to social stimuli, leading to interference. Thus, these stable individual differences are clear in our data, and interacted in some cases, although ambiguously, with the conditioning procedure (e.g., anxious individuals responded more quickly to all words in the variable ratio condition).

Bridging a gap between behaviorist and attachment theorists that persisted for many decades, this study provides some of the first direct support that negative reinforcement processes can influence attachment attitudes and motivation. The quality of attitudes and motivations may emerge, at least in part, out of the schedule of reinforcement in predictable ways consistent with extant models (Beckes & Coan, 2015; Mikulincer et al., 2003; Johnson et al., 2013). These ideas must be tested in increasingly ecologically valid contexts before acceptance of any application value, but if it generalizes, such information could help couples with attachment-related conflicts (cf. Johnson et al., 2013) and be a boon to psychological research and practice (cf. Mikulincer & Shaver, 2007).

Test Type	Specific Prediction	Fit Prediction by p Cutoff	Fit Prediction by Pattern	Tails^a	p Value
Response times
Omnibus three-way interaction	Conditioning × Word Type × Valence interaction	Yes	Yes	N/A	.031
Nested two-way interactions
Attachment words	Conditioning × Valence interaction	Yes	Yes	N/A	.028
Nonattachment words	No Conditioning × Valence interaction (null)	Yes	Yes	N/A	.506
Nested paired-comparisons
Positive attachment words	Variable ratio < continuous	No	Yes	1	.06
Variable ratio < neutral	Yes	Yes	1	.0025
Continuous < neutral	No	Yes	1	.108
Negative attachment words	Variable ratio ≠continuous	No	N/A	2	.48
Variable ratio ≠ neutral	No	N/A	2	.417
Continuous > neutral	No	Yes	1	.464
Explicit evaluations
Omnibus ANOVA	Conditioning main effect	Yes	Yes	N/A	<.001
Nested paired-comparisons	Variable ratio < continuous	Yes	Yes	1	<.001
Variable ratio < neutral	Yes	Yes	1	<.001
Continuous > neutral	Yes	Yes	1	<.001
P1 ERPs
Omnibus ANOVA	Conditioning main effect	Yes	Yes	N/A	.041
Conditioning × Hemisphere interaction	No	Yes	N/A	.607
Nested paired-comparisons	Variable ratio > continuous	Yes	Yes	1	.035
Variable ratio > neutral	Yes	Yes	1	.0085
Continuous > neutral	No	Yes	1	.237

Schedule	Attachment Words	Nonattachment Words
Variable ratio	587.78	3.69	609.36	5.25	619.14	6.02	616.03	6.47
Continuous	597.84	4.69	604.32	3.80	620.76	6.20	620.45	6.15
Neutral	605.63	5.28	603.78	4.22	613.92	4.28	623.26	6.81

Footnotes

Authors’ Note

Beckes developed the hypotheses and study design. Simons, Lewis, Le, and Edwards collected data under the supervision of Beckes. All authors assisted in data analysis led by Beckes. All authors contributed sections to the first draft of the article. Beckes edited the final draft of the article. All authors have approved the final manuscript for submission.

Acknowledgments

We thank Dallas Garrison, Casey Grage, Wendy Hasse, Brett Hoffman, Kylie McKinney, Victor Medina, Marina Palakeel, and Clara Tostovarsnik for their help in collecting data.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Supplemental Material

The data supplements are available in the online version of the article.

Notes

References

Anselme

Robinson

M. J. F.

Berridge

K. C.

(2013). Reward uncertainty enhances incentive salience attribution. Behavioral and Brain Research, 238, 53–61.

Baumeister

R. F.

Leary

M. R.

(1995). The need to belong: Desire for interpersonal attachments as a fundamental human motivation. Psychological Bulletin, 117, 497–529.

Beckes

Coan

J. A.

(2011). Social baseline theory: The role of social proximity in emotion and economy of action. Personality and Social Psychology Compass, 5, 976–988.

Beckes

Coan

J. A.

(2013). Toward an integrative neuroscience of relationships. In Simpson

J. A.

Campbell

(Eds.), The Oxford handbook of close relationships (pp. 684–710). New York, NY: Oxford University Press.

Beckes

Coan

J. A.

(2014). Relationship neuroscience. In Simpson

J. A.

Dovidio

(Eds.), The handbook of personality and social psychology, Volume 3: Interpersonal relationships and group processes (pp. 119–149).Washington, DC: American Psychological Association Press.

Beckes

Coan

J. A.

(2015). The distress-relief dynamic in attachment bonds. In Hazan

Zayas

(Eds.), Bases of adult attachment: Linking brain, mind, and behavior (pp. 11–33). New York: Springer Press.

Beckes

Coan

J. A.

Morris

J. P.

(2013). Implicit conditioning of faces via the social regulation of emotion: ERP evidence of early attentional biases for security conditioned faces. Psychophysiology, 50, 734–742.

Beckes

Ijzerman

Tops

(2015). Toward a radically embodied neuroscience of attachment and relationships. Frontiers in Human Neuroscience. doi:10.3389/fnhum.2015.00266

Beckes

Simpson

J. A.

Erickson

A. B.

(2010). Of snakes and succor: Learning secure attachment associations with novel faces via negative stimulus pairings. Psychological Science, 21, 721–728.

10.

Bowlby

(1982). Attachment and loss, volume I: Attachment. New York, NY: Basic Books. (Original work published 1969)

11.

Bradley

M. M.

Lang

P. J.

(1994). Measuring emotion: The Self-Assessment Manikin and the semantic differential. Journal of Behavior Therapy and Experimental Psychiatry, 25, 49–59. doi:10.1016/0005-7916(94)90063-9

12.

BrainVision, LLC. (2016). ActiChamp [Brochure]. Retrieved from www.brainvision.com/actichamp.html

13.

Brennan

K. A.

Clark

C. L.

Shaver

P. R.

(1998). Self-report measurement of adult romantic attachment: An integrative overview. In Simpson

J. A.

Rholes

W. S.

(Eds.), Attachment theory and close relationships (pp. 46–76). New York: Guilford Press.

14.

Cho

H.-C.

Abe

(2013). Is two-tailed testing for directional research hypotheses legitimate? Journal of Business Research, 6, 1261–1266.

15.

Cousineau

(2005). Confidence intervals in within-subjects designs: A simpler solution to Loftus and Masson’s method. Tutorial in Quantitative Methods for Psychology, 1, 42–45.

16.

Cunningham

W. A.

Van Bavel

J. J.

Arbuckle

N. L.

Packer

D. J.

Waggoner

A. S.

(2012). Rapid social perception is flexible: Approach and avoidance motivational states shape P100 responses to other-race faces. Frontiers in Human Neuroscience, 6. doi:10.3389/fnhum.2012.00140

17.

Dan

Raz

(2012). Adult attachment and emotional processing biases: An event related potentials (ERPs) study. Biological Psychology, 91, 212–220.

18.

Delorme

Makeig

(2004). EEGLAB: An open source toolbox for analysis of single-trial EEG dynamics. Journal of Neuroscience Methods, 134, 9–21.

19.

Fraley

R. C.

Vicary

A. M.

Brumbaugh

C. C.

Roisman

G. I.

(2011). Patterns of stability in adult attachment: An empirical test of two models of continuity and change. Journal of Personality and Social Psychology, 101, 974–992.

20.

Fraley

R. C.

Waller

N. G.

Brennan

K. A.

(2000). An item-response theory analysis of self-report measures of adult attachment. Journal of Personality and Social Psychology, 78, 350–365.

21.

Harlow

(1958). The nature of love. American Psychologist, 13, 673–685.

22.

Hazan

Shaver

(1987). Romantic love conceptualized as an attachment process. Journal of Personality and Social Psychology, 5, 511–524.

23.

Holt-Lunstad

Smith

T. B.

Layton

J. B.

(2010). Social relationships and mortality risk: A meta-review. PLoS Medicine, 7, e1000316. doi:10.1371/journal.pmed.1000316

24.

IBM Corp. (2013). SPSS Statistics for Macintosh (Version 22). Armonk, NY: IBM Corp.

25.

Jeffreys

(1961). Theory of probability (3rd ed.). Oxford, England: Oxford University Press.

26.

Johnson

S. M.

Moser

M. B.

Beckes

Smith

Dalgleish

Halchuck

… Coan

J. A.

(2013). Soothing the threatened brain: Leveraging contact comfort with emotionally focused therapy. PLoS One, 8, e79314. doi:10.1371/journal.pone.0079314

27.

Klohnen

E. C.

Weller

J. A.

Luo

Chloe

(2005). Organization and predictive power of general and relationship-specific attachment models: One for all, and all for one? Personality and Social Psychology Bulletin, 31, 1665–1682.

28.

Markley

Luck

Lopez-Calderon

(2011). ERPLAB toolbox user’s manual. Retrieved June 2012, from http://erpinfo.org/erplab/erplab-documentation/manual

29.

Mikulincer

Gillath

Shaver

P. R.

(2002). Activation of the attachment system in adulthood: Threat-related primes increase the accessibility of mental representations of attachment figures. Journal of Personality and Social Psychology, 83, 881–895.

30.

Mikulincer

Shaver

P. R.

(2005). Attachment theory and emotions in close relationships: Exploring the attachment-related dynamics of emotional reactions to relational events. Personal Relationships, 12, 149–168.

31.

Mikulincer

Shaver

P. R.

(2007). Boosting attachment security to promote mental health, pro-social values, and inter-group tolerance. Psychological Inquiry, 18, 139–165.

32.

Mikulincer

Shaver

P. R.

Pereg

(2003). Attachment theory and affect regulation: The dynamics, development, and cognitive consequences of attachment-related strategies. Motivation and Emotion, 27, 77–102.

33.

Simpson

J. A.

Collins

W. A.

Tran

Haydon

K. C.

(2007). Attachment and the experience and expression of emotion in romantic relationships: A developmental perspective. Journal of Personality and Social Psychology, 92, 355–367.

34.

Skinner

B. F.

(1956). A case history in scientific method. American Psychologist, 11, 221–233.

35.

Zayas

Shoda

(2005). Do automatic reactions elicited by thoughts of romantic partner, mother, and self relate to adult romantic attachment? Personality and Social Psychology Bulletin, 31, 1011–1025.

36.

Zayas

Shoda

(2014). Love you? Hate you? Maybe its both: Evidence that significant others trigger bivalent priming. Social Psychological and Personality Science, 6, 56–64.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.15 MB

Schedule	Attachment Words				Nonattachment Words
	Positive		Negative		Positive		Negative
	M	SE	M	SE	M	SE	M	SE
Variable ratio	587.78	3.69	609.36	5.25	619.14	6.02	616.03	6.47
Continuous	597.84	4.69	604.32	3.80	620.76	6.20	620.45	6.15
Neutral	605.63	5.28	603.78	4.22	613.92	4.28	623.26	6.81