Social support for healthy behaviors: Scale psychometrics and prediction of weight loss among women in a behavioral program

Abstract

Social support could be a powerful weight-loss treatment moderator or mediator but is rarely assessed. We assessed the psychometric properties, initial levels, and predictive validity of a measure of perceived social support and sabotage from friends and family for healthy eating and physical activity (eight subscales). Overweight/obese women randomized to one of two 6-month, group-based behavioral weight-loss programs (N=267; mean BMI 32.1±3.5; 66.3% White) completed subscales at baseline, and weight loss was assessed at 6 months. Internal consistency, discriminant validity, and content validity were excellent for support subscales and adequate for sabotage subscales; qualitative responses revealed novel deliberate instances not reflected in current sabotage items. Most women (>75%) “never” or “rarely” experienced support from friends or family. Using non-parametric classification methods, we identified two subscales—support from friends for healthy eating and support from family for physical activity—that predicted three clinically meaningful subgroups who ranged in likelihood of losing ≥5% of initial weight at 6 months. Women who “never” experienced family support were least likely to lose weight (45.7% lost weight) whereas women who experienced both frequent friend and family support were more likely to lose weight (71.6% lost weight). Paradoxically, women who “never” experienced friend support were most likely to lose weight (80.0% lost weight), perhaps because the group-based programs provided support lacking from friendships. Psychometrics for support subscales were excellent; initial support was rare; and the differential roles of friend versus family support could inform future targeted weight-loss interventions to subgroups at risk.

Social support is considered a key component of behavioral weight-loss programs (1, 2). Given that social contexts can both help and hinder weight-loss efforts (3), programs frequently teach strategies to elicit support and manage sabotage from friends and family. Whereas observational studies and randomized trials have examined the role and involvement of support partners in weight-loss programs (4-11), the constructs of perceived support and sabotage per se are rarely assessed within weight-loss trials (6, 12). This hampers subsequent intervention research investigating whether support and sabotage may be moderators or mediators of treatment outcomes (13) and how future trials can target interventions to at-risk subgroups (14).

A questionnaire measuring perceived social support and sabotage from friends and family for healthy eating and physical activity behaviors (eight subscales total) was originally developed by Sallis and colleagues (15) and subsequently revised by Ball and Crawford (16). The original items, constructed from qualitative interviews, were factor analyzed, and resulting subscales were validated using self-reported food frequency and activity questionnaires among a small sample of primarily White college students attempting dietary and/or activity changes (15). In the Sallis version (15), the number and content of items (based on eigenvalues) varied across subscales making it difficult to compare the role of support and sabotage by type of health behavior (e.g., healthy eating or activity) or source of support (e.g., friend or family). For example, there were seven items assessing sabotage from family for healthy eating whereas there were no subscale items assessing sabotage from friends for physical activity. In the Ball and Crawford revision (16), the number and content of items were made consistent across subscales (with some original items deleted and some new items added) and the internal consistency of the subscales was examined among a sample of 790 young women. To date, the support and sabotage subscales of either version have not been validated among overweight/obese adults participating in weight-loss programs or among samples of adults diverse in age and ethnicity.

In the current study, we assessed the psychometric properties of the Ball and Crawford (16) subscales including internal consistency, discriminant validity (to date not examined), and content validity; and assessed initial levels of social support and sabotage among overweight/obese women starting one of two 6-month behavioral weight-loss programs in a randomized controlled trial. We also assessed predictive validity, i.e., whether subscales predicted subgroups who were more successful at losing weight after the 6-month programs.

Methods

Participants and procedure

Analyses for the current psychometric study were based on data from an 18-month randomized controlled behavioral weight management trial (17). In the trial, overweight/obese women (N=267) were recruited from Northern California communities and randomized to one of two 6-month, group-based behavioral weight-loss programs. For assessments, they completed online questionnaires and attended clinic visits at baseline, 6, 12, and 18 months. Eligibility criteria included: age ≥21 years; free of heart disease, diabetes, and other chronic health conditions; able to participate in physical activity (18); body mass index (BMI) 27-40 kg/m²; free of binge eating disorder or bulimic compensatory symptoms (19); access to the Internet; and interest in attending classes offered in English. The trial was approved by the Stanford University Institutional Review Board and all participants provided informed consent.

For the current psychometric study, we used data from the online questionnaires at baseline and from the clinic visits at baseline and 6 months. Retention was excellent; only 4.9% (n=13) of 267 randomized participants did not return for their 6-month clinic visit with no differences by type of weight-loss program (p=0.81). Values for the missing 6-month weights were imputed using the baseline carried forward approach (2).

Measures

Demographics

Age in years, education status (categories ranging from no schooling to professional degree), race/ethnicity (American Indian/Alaska Native, Asian, Black/African American, Latina/Hispanic, Native Hawaiian/Pacific Islander, and/or White), marital status (married, living with a partner or significant other, divorced/separated, widowed, or single), number of children, and prior participation in a formal weight-loss program (yes or no) were measured with validated single-item questions (20). Using standardized protocols (21), body weight was measured on a standard beam balance with participants in light clothing and without shoes; height was measured using a stadiometer. BMI was calculated as weight in kilograms divided by height in meters squared.

Social support and sabotage for healthy eating and physical activity subscales

Perceived support and sabotage from friends and family for healthy eating and physical activity (eight subscales) were measured with the Ball and Crawford 36-item version (16). Participants rated how frequently in the past month friends (or family) engaged in particular activities or said particular statements, using a 4-point Likert scale with response options labeled never/not applicable, rarely, sometimes, and often (items available upon request). The four support subscales each had six items including two behavioral items (e.g., “Ate healthy foods with me”) and four verbal items (e.g., ”Complimented me on my eating habits”). The four sabotage subscales each had three items including two behavioral items (e.g., ”Ate high-fat or unhealthy foods in front of me”) and one verbal item (e.g., ”Refused to eat healthy or low-fat foods with me”). In the Sallis version (15), internal consistency (Cronbach’s α=0.61–0.91) and test-retest reliability (rs=0.55–0.86) ranged widely, likewise for the Ball and Crawford version (Cronbach’s α=0.58–0.81; 16).

In contrast to prior work (15, 16) and to improve interpretation, we did not reverse score items on the sabotage subscales and we created each subscale score by summing responses and dividing by the number of items. Thus, higher scores reflect more frequent support or sabotage, and mean subscale scores correspond directly to response options. Also, we used a shorter time frame, “in the past month” versus “in the past three months” or “during the past 2 years” (15, 16).

General social support and strain subscales

To evaluate discriminant validity, we assessed general (non-specific) supportive and strained interactions with friends, family, and spouse (6 subscales, 28 items total; 22, 23). Participants rated either how much or how frequently others respond to them, using a 4-point Likert scale with options labeled not at all/never, a little/rarely, some/sometimes, and a lot/often (not applicable was a fifth option and not included in subscale means). General support scales had four items each for family (or friends or spouses), (e.g., ”how much can you rely on them for help if you have a serious problem?”) as well as two additional items for spouses (e.g., ”how much does he or she appreciate you?”). General strain subscales also had four items each for family (or friends or spouses), (e.g., ”how often do they make too many demands on you?”) and two additional items for spouses (e.g., ”how often does he or she argue with you?”). The general subscales have excellent psychometrics (e.g., Cronbach’s α=0.79–0.88) among large population-based samples (22, 23).

Social support for healthy behaviors (qualitative question)

An open-ended question was included to elicit themes which may or may not have been represented in existing subscale items: ”What, if anything, would you like to add about the social support in your life with regard to your eating habits, physical activity habits, or weight?” This question was included in the online questionnaires for the latter half of study participants.

Statistical analyses

Internal consistency for each of the social support and sabotage subscales was examined with Cronbach’s α and discriminant validity was examined with Spearman correlations (24, 25). Content validity was evaluated by two raters (DS and SM) coding responses to the open-ended question using inductive (i.e., deriving codes from emergent themes) and deductive methods (i.e., deriving codes from existing scale items; 26). Using the codebook, a third rater (JM), blind to aims, coded responses. Inter-rater reliability was excellent (defined as ≥.80; 27), κ=0.84.

To examine predictive validity, we used signal detection to determine whether the support and sabotage subscales predicted subgroups of women more likely to lose ≥5% of initial body weight at 6 months. Weight loss of ≥5% was chosen given consensus that this magnitude is clinically significant and related to cardiovascular disease risk factor reduction (28). Signal detection is a non-parametric risk classification method that uses empirically driven recursive partitioning to produce a series of “and/or” (Boolean) rules that identify subgroups more or less likely to have a binary outcome (14). In contrast to logistic regression, signal detection has the ability to identify subgroups that are homogeneous in both predictor variables and the outcome (thus, making it easier to subsequently design interventions targeted to different subgroups) as well as the ability to identify subgroups that are similar in the likelihood of the outcome but may have very different descriptive profiles (thus, suggesting the need to design interventions tailored to different subgroups; 14). In this study, signal detection identified variables which optimally predicted subgroups with and without the binary outcome by evaluating incremental cut points for each predictor variable across all predictor variables according to a clinical criterion. Here, we chose efficiency which equally weights false positives and false negatives as there was no a priori reason in this context to weight one over the other (14). The cut point with the highest efficiency was tested for significance using a 2 × 2 chi-square test at a chosen significance level (i.e., p<.01). After the most efficient cut point was identified, it was used to divide the group into two subgroups and the iterative process was repeated in each of the two subgroups, and then in each generation of subgroups until there were too few people in a subgroup (n<10) or the most efficient cutoff was non-significant (14). To interpret the role of predictor variables (e.g., the 8 subscales) in characterizing the final subgroups (14), we then examined how the final subgroups differed on both predictor variables and profile variables (e.g., demographics).

Results

Demographics

Women (N=267) were middle-aged (age range 21.6-75.3 years) and most had a college degree (Table 1). A third were non-White women including Latina/Hispanic (10.5%, n=28), multiethnic (i.e., two or more races/ethnicities, 10.1%, n=27), Asian (9.4%, n=25), Black/African American (3.0%, n=8) and Native Hawaiian/Pacific Islander women (0.7%, n=2) reflecting the region’s ethnic/racial diversity (29). Two-thirds of the women were married or living with a partner, had at least one child, had a BMI ≥30, and had participated in a prior formal weight-loss program such as Weight Watchers.

Table 1. Baseline characteristics (N=267).

Characteristic	Mean ± SD or proportion/n
Demographics
Age, years	48.4 ± 10.8
College degree (%)	67.0% (n=179)
Non-White racial/ethnic background (%)	33.7% (n=90)
Married/living with partner (%)	68.9% (n=184)
Had at least one child* (%)	67.4% (n=174)
Initial body mass index	32.1 ± 3.5
Initial body mass index ≥30 (%)	65.2 (n=174)
Participated in prior formal program (%)	67.8% (n=181)
Social support (6-item subscales) **
From friends for healthy eating (α=0.83)	2.0 ± 0.7
From friends for physical activity (α=0.85)	2.0 ± 0.7
From family for healthy eating (α=0.82)	2.3 ± 0.8
From family for physical activity (α=0.84)	2.3 ± 0.8
Sabotage (3-item subscales) **
From friends for healthy eating (α=0.70)	2.1 ± 0.7
From friends for physical activity (α=0.66)	1.9 ± 0.7
From family for healthy eating (α=0.73)	2.2 ± 0.8
From family for physical activity (α=0.61)	2.3 ± 0.8

^*Based on smaller sample size due to missing data (n=258).

^**To improve interpretation, we calculated a mean score for each subscale (i.e., summed item responses and divided by the number of items). Thus, higher mean scores reflect more frequent support (or sabotage) and correspond directly to response options (1=Never, 2=Rarely, 3=Sometimes, 4=Often).

Internal consistency

Internal consistency (Cronbach’s α) was excellent (defined as ≥0.80; 24, 25) for the four social support for healthy behavior subscales but only poor (defined as <0.70) to adequate (defined as 0.70-0.79) for the four original 3-item sabotage subscales (Table 1). The sabotage item regarding refusal to participate in healthy behaviors was infrequently endorsed (<21%). Removing this item improved internal consistency for all but the sabotage from family for physical activity subscale (0.86, 0.78, 0.78, 0.69, respectively).

Initial social support and sabotage

Initial mean levels of social support for healthy behaviors subscales were low (Table 1). Indeed, most women “never” or “rarely” experienced support from friends for healthy eating (90.3%) or physical activity (87.6%), nor from family for healthy eating (77.9%) or physical activity (77.2%). Initial levels of sabotage subscales were similar to support subscales.

Discriminant validity

We examined discriminant validity in two sets of analyses. In the first set of analyses, we examined whether the social support for healthy behaviors subscales were related to the general support subscales. Although internal consistency for general support subscales was excellent (α=0.82-0.92) and initial levels of general support were high (M=3.3±0.6-3.5±0.6), the social support for healthy behaviors subscales were not highly related to the general social support subscales (rs=−0.10–0.27), demonstrating discriminant validity. Likewise, we examined whether the sabotage for healthy behaviors subscales were related to the general strain subscales. Although internal consistency for the general strain subscales was excellent (α=0.76-0.88) and initial levels of general strain were low (M=1.9±0.5-2.4±0.7), the sabotage for healthy behaviors subscales were not highly related to the general strain subscales (rs=0.05-0.32).

In the second set of analyses for discriminant validity, we examined whether the eight subscales for support and sabotage for healthy behaviors were related to one another by kind of support (i.e., support or sabotage), source of support (i.e., friends or family), or type of health behavior (i.e., healthy eating or activity). Subscales were not highly related by kind or source of support (rs=−0.03–0.37), demonstrating discriminant validity. However, subscales were related by type of health behavior for each source of support, i.e., the support for healthy eating subscale was related to the support for physical activity subscale for friends (r=0.53) and for family (r=0.67). In addition, the sabotage for healthy eating subscale was related to the sabotage for physical activity subscale for friends (r=0.44) and for family (r=0.49).

Content validity

Qualitative analysis of the responses to the open-ended question revealed four sets of themes. First, the open-ended question elicited themes about support and sabotage already reflected in existing subscale items, supporting the content validity of the Ball and Crawford version (sample statements appear in Table 2). Second, the open-ended question elicited themes about sabotage reflected in items of the Sallis version subsequently dropped in the Ball and Crawford revision (e.g., “criticized me for my eating habits”). Third, the open-ended question elicited themes about support and sabotage not reflected in either version. Most notable were qualitative instances of deliberate attempts to discourage healthy lifestyle behaviors, using verbs such as “tempt,” “undermine,” and “sabotage.” These were distinguishable from existing sabotage items (e.g., “spent time being inactive around me”) that may not be deliberate per se. Fourth, the question elicited other themes not reflected in either version, including desire for a workout buddy, desire for less support, recognition of being the one responsible, feeling isolated, and mistreatment from others because of weight.

Table 2. Qualitative Themes and Quotes Regarding Social Support for Healthy Behaviors (n=113).

Theme	Sample quote
I. Support and sabotage themes reflected in Ball & Crawford (2006) items
Social support
Participated in physical activity with me	I walk with my husband mostly every day, we walk our dogs.
Offered to eat healthy foods with me	Well my friend just joined a gym and she is working out and eating good. She asked me to join and walk together at work and eat more health[y].
Offered to participate in physical activity with me	My husband is very support[ive] about offering to go to the gym with me.
Complimented me on my physical activity habits	I have adult children who think I am doing very well with my physical activity.
Encouraged me to be more physically active	My daughter is especially supportive of healthy eating habits and encourages me to exercise.
Sabotage
Offered me high-fat or unhealthy foods	Some other family members, though, do try to sabotage my efforts by piling mounds of food on my plate and won’t take no for an answer.
Refused to eat healthy with me	My family does not/will not eat healthy foods. It makes it very hard for me to watch what I eat.
Spent time being inactive around me	My friends…are conscious of healthful eating and great cooks, but mostly sedentary, fond of sugar and alcohol, and talk about being active more than actually being so, including me!
II. Sabotage themes reflected in Sallis (1987) deleted items
Sabotage
Brought home high-fat or unhealthy foods	My husband is very thin and thinks it is fine that he bring high fat and high sugar items home regularly….This is very frustrating.
Criticized me for my eating habits	I wish my boyfriend…didn’t put us in so many social situations where there is alcohol….He makes me feel guilty about not drinki[n]g with everyone.
Criticized me for my physical activity habits	[I have] a husband who criticizes my weight, inactivity.
III. Support and sabotage themes not reflected in existing items
Social support
Ate healthy or low-fat foods in front of me	My husband eats healthfully and cooks good food. He also exercises most days, and encourages everyone to eat like him.
Spent time being physically active around me	My daughter & husband are both very active people. My daughter bikes to work and exercises regularly and my husband bikes for pleasure.
Sabotage
Undermined my efforts to [eat healthy or low-fat  foods/be physically active]	My boyfriend often brings home high calorie and high fat foods and tempts me with them. This is especially hard for me to resist when I have just gotten home from work and I am very tired and hungry.
IV. Other themes not reflected in existing items
I wish my [family/friends] would [eat healthy or  low-fat foods/participate in physical activity] with  me	I would like more support from my friends and family. Like walking with my co-worker at lunch time. Hiking on the weekend with my family.
I don’t want my [family/friends] to be involved in  my efforts to [eat healthy or low-fat foods/be  physically active]	I don’t want my family/friends (social support) to be involved at all. I definitely don’t want to be reminded to do something (e.g., exercise, eat better, etc.), especially when I’m struggling with it. I already feel badly about myself, reminders only make me feel worse, and seem unnecessary.
I am the only one responsible for my choices [to eat  healthy or low-fat foods/be physically active]	I am the one who has to increase my desire to be more active. I have no excuses.
Social isolation	None of my family live within 2000 miles of me. My best friends live out of the area, so I don’t have a strong support group.
Weight-related mistreatment	My family just puts me down because of my weight but they don’t give any positive support.

Predictive validity

For the signal detection results, we first describe which predictor variables were identified at each step of the iterative process (Figure 1), and then examine how the final subgroups differed on predictor and profile variables (Table 3). Of the predictor variables (i.e., the eight subscales and type of weight-loss program), signal detection identified two subscales—support from friends for healthy eating and support from family for physical activity—which predicted three clinically meaningful subgroups of women who ranged widely in the likelihood of losing ≥5% of initial weight at 6 months (45.7–80.0%; Figure 1). Among the full sample (N=267), frequency of support from friends for healthy eating was identified as the first optimally efficient predictor, κ=0.15, χ2(1, N=267)=7.5, p<.01, and divided the sample into two groups using a cut point of <2. Among the group of women who experienced less frequent support from friends (value <2, n=130), no additional predictors were found to be optimally efficient. Among the group of women who experienced more frequent support from friends (value ≥2, n=137), frequency of support from family for physical activity was the next optimally efficient predictor, κ=0.23, χ2(1, 137)=7.7, p<.01, and divided that group into two subgroups using a cut point of ≥2. No additional predictors were found to be optimally efficient in either of the remaining subgroups.

Figure 1. Signal Detection Subgroups.

Table 3. Outcome, Predictor, and Profile Variables by Signal Detection Subgroup.

	Signal detection subgroups
	More frequent friend support		Infrequent friend support
	Infrequent family support	Frequent family support		p
Sample size	(n=35)	(n=102)	(n=130)
Outcome variable
Lost ≥5% of initial weight at 6 months (%)	45.7% (n=16)	71.6% (n=73)	80.0% (n=104)	.0003
Predictor variables
Social support
From friends for healthy eating	2.5 ± 0.4 a	2.5 ± 0.4 b	1.4 ± 0.3 a,b	<.0001
From friends for physical activity	2.3 ± 0.8 a	2.3 ± 0.6 b	1.7 ± 0.7 a,b	<.0001
From family for healthy eating	1.9 ± 0.7 a	2.8 ± 0.6 a,b	2.0 ± 0.7 b	<.0001
From family for physical activity	1.5 ± 0.3 a,b	2.8 ± 0.5 a,c	2.0 ± 0.8 b,c	<.0001
Sabotage
From friends for healthy eating	2.0 ± 0.7 a	2.3 ± 0.6 a,b	2.0 ± 0.7 b	.003
From friends for physical activity	1.9 ± 0.7	2.1 ± 0.6 a	1.7 ± 0.7 a	<.0001
From family for healthy eating	2.2 ± 0.9	2.2 ± 0.6	2.3 ± 0.8	.77
From family for physical activity	2.2 ± 0.9	2.4 ± 0.6	2.2 ± 0.8	.14
Profile variables
Demographics
Age, years	47.6 ± 12.2	47.7 ± 10.9	49.1 ± 10.4	.56
College degree (%)	68.6% (n=24)	64.7% (n=66)	68.5% (n=89)	.82
Non-White ethnic background (%)	42.9% (n=15)	44.1% (n=45)	23.1% (n=30)	.002
Married/living with partner (%)	45.7% (n=16)	74.5% (n=76)	70.8% (n=92)	.005
Has at least one child (%)*	52.9% (n=18)	69.0% (n=69)	70.2% (n=87)	.15
Initial body mass index	31.1 ± 3.1	31.9 ± 3.5	32.5 ± 3.5	.07
Initial body mass index ≥30 (%)	51.4% (n=18)	64.7% (n=66)	69.2% (n=90)	.14
Participated in prior formal program (%)	60.0% (n=21)	65.7% (n=67)	71.5% (n=93)	.36
Weight change at 6 months
Percent weight change	−6.2 ± 6.7 a,b	−8.7 ± 6.7 a	−9.6 ± 5.8 b	.02
Percent weight change (median)	−4.6	−8.4	−9.8
Body mass index change	−1.9 ± 2.1 a,b	−2.8 ± 2.2 a	−3.1 ± 1.9 b	.01
Body mass index change (median)	−1.3	−2.7	−3.1
Weight change, lbs	−11.1 ± 12.2 a,b	−16.2 ± 12.6 a	−18.4 ± 11.6 b	.006
Weight change, lbs (median)	−8.0	−16.6	−18.4

Note. Values are expressed as mean ± standard deviation unless otherwise noted. Across columns, values that share a letter are significantly different, p<0.05.

^*Values for this item were based on a smaller sample size (n=258) due to missing data.

For the three final subgroups identified in Figure 1, descriptive information is provided for outcome, predictor, and profile variables by subgroup in Table 3. In the first subgroup (left box of Figure 1), a small group of women infrequently experienced support from family for physical activity (13.1% of sample) and only 45.7% of them lost weight. They were less likely to be married/living with a partner but as likely to have at least one child as other subgroups (left column of Table 3). In the second subgroup (middle box of Figure 1), women more frequently experienced support from family for physical activity (38.2% of sample) and 71.6% lost weight. These women experienced more frequent support from both friends and family for healthy behaviors, but also more sabotage than other subgroups (middle column of Table 3). In the third subgroup (right box of Figure 1), a large group of women infrequently experienced support from friends for healthy eating and physical activity (48.7% of sample) and 80.0% lost weight. These women were more likely to be White than other subgroups (right column of Table 3).

Discussion

Social support for lifestyle behaviors such as healthy eating and physical activity is an essential component of behavioral weight-loss programs yet measures of these constructs have rarely been included as assessments in weight-loss intervention trials. To our knowledge, this study is the first to examine the psychometric properties for a measure of perceived social support and sabotage for healthy behaviors among overweight/obese women in a weight-loss intervention trial. Internal consistency, discriminant validity, and content validity were excellent for support subscales and adequate for sabotage subscales.

Given the psychometric results, we make some conceptual and methodological recommendations. Conceptually, sabotage subscales need further work. The term “sabotage,” as defined by the Oxford English Dictionary, denotes a deliberate action (30). The qualitative analysis revealed additional deliberate instances (e.g., being criticized or explicitly tempted with food). Thus, developing additional items may improve content validity and psychometrics. Methodologically, scoring adjustments should be made. First, using means rather than totals for subscale scores to directly reflect response labels will simplify interpretation. Second, modifying the response options for subscales will improve reliability. By adding almost always as a fifth option to avoid a built-in ceiling effect (and coding response options from one to five), each option would continue to have a label (increasing reliability) with matching labels at each end (almost never, rarely, sometimes, often, and almost always; 31). Future research can examine whether subscales are responsive to change, a precursor to investigating whether these social constructs are powerful mediators of intervention outcomes.

Overall, despite reporting relatively high levels of general social support in their relationships with friends, family and spouses, this sample of overweight/obese women starting a behavioral weight-loss program experienced infrequent social support for healthy eating and physical activity behaviors. The low frequencies are more dramatic considering the potential ceiling effect of the highest response option (i.e., “often” rather than “almost always”). Of particular interest is the potentially powerful role of friends (not just family), which is consistent with intriguing epidemiological results documenting that individuals were 57% more likely to become obese if they had a friend who became obese within the same time period (32). Future research is needed that assesses the complexities of friends and family members’ abilities to provide social support for healthy behaviors given their own weight status and the composition of the household (33).

Given that women reported infrequently experiencing support for healthy behaviors, it would be worthwhile to evaluate whether the women’s perceptions of support from their friends and family are consistent with the friends and family’s own perceptions of support for the women. For instance, examining the consistencies (or discrepancies) between the perceptions of adolescents and their sexual partners has yielded useful insights about high-risk sexual behaviors (e.g., 34). Such a methodological approach may be equally useful here. In light of the broad reach of the obesity epidemic in the United States, women may be surrounded by friends and family who do not value or practice healthy lifestyle behaviors (35). Alternatively, women could be (mis)interpreting the lack of verbal and behavioral responses in an overly negative manner (12). Friends and family may want to be supportive but hesitate to say or do the “wrong” thing and, thus, do nothing, especially as most women have “been here before” such as participating in a prior program. Women could also be “self-handicapping” (even unintentionally) by setting up a preemptive rationale for not losing weight. Women must successfully navigate the overwhelmingly obesogenic environment surrounding them. However, despite the importance of environmental variables, individual perceptions often drive behavior change, and thus, perceptions of support and sabotage across social realms remain important constructs for continued investigation.

The signal detection results provide initial evidence for the predictive validity of the support subscales given that three distinct subgroups were identified which differed in clinically relevant weight loss. The heterogeneity in initial support may help explain inconsistent findings in the past literature regarding the role of friends and family as support partners in weight loss trials (4-9, 11). Some past trials have shown that individuals who participate with partners lose more weight than individuals who participate in a program alone (e.g., 5, 9) especially if partners lose weight (8). However, individuals in these trials are typically not randomized to the two programs; instead they self-select to participate with a partner or alone. Past trials likely recruited individuals into each program with different levels of initial support, i.e., individuals who participate alone may not have partners or have unsupportive partners. More recently, two trials have recruited individuals with willing support partners and then randomized them to participate with a partner or alone (9, 11). Yet, these trials likely sampled individuals with high levels of initial support and missed individuals with low levels. Ideally, systematically assessing initial levels of support would provide useful comparison data about study samples across intervention trials.

The signal detection results, especially the heterogeneity of support from friends, could inform the design of new interventions targeted for each subgroup. Whereas we speculate below about what these targeted interventions might look like, we clearly recognize the necessity of first conducting additional rich qualitative and quantitative research about each of the subgroups to further inform how best to design such interventions. Furthermore, such targeted interventions would need to be rigorously tested in future randomized trials with sufficient statistical power to determine if proposed mediators indeed change in the targeted intervention compared to a control or comparison intervention evident by differential increases in relevant support subscales by intervention group (13).

In the first subgroup (left box of Figure 1), women who “never” experienced family support were least likely to lose weight at 6 months—only 45.7% did so. Interestingly, these women, who were less likely to be married/living with a partner and included single parents, reported a level of support from friends comparable to the second subgroup which experienced the most frequent support from both friends and family. Thus, an intervention targeted for the first subgroup could focus on nurturing existing positive friendships rather than developing new friendships or increasing support from family members. Such an intervention builds on what is already working well in these women’s busy lives, although care would need to be taken that the intervention did not exacerbate conflict or undermine any existing support from family members. There are numerous creative ways to nurture friend support, such as using social media. For instance, people in different geographical locations and even time zones can talk to one another while exercising together using hands-free mobile phones. Indeed, technology exists that motivates people to maintain their walking pace while talking to another person—each person’s heartbeat is measured, and if one person slows down (with corresponding lower heart rate), the technology lowers the volume of the slower person’s voice on the phone, thus encouraging the slower person to return to his/her original pace to continue the conversation (36).

In the second subgroup (middle box of Figure 1), and perhaps as expected, women who experienced the most frequent support from friends and family were very successful at losing weight at 6 months—71.6% did so. However, they also experienced the highest sabotage, especially from friends for healthy eating. These data are consistent with a small qualitative study using social network approaches in which 40% of participants identified family members as both the most and least helpful (37), and argue for simultaneously considering the relative influences of both forms of social interaction, such as conflicting forces or support as a buffer from more negative (but perhaps overall infrequent) interactions. Thus, an intervention targeted for women in the second subgroup could focus on systematically examining their social network to better manage relationships involving sabotage. However, it is also important to note that although these women experienced the highest support relative to other subgroups, on average they only “sometimes” received support, leaving room for considerable improvement.

In the third subgroup (right box of Figure 1), women who “never” experienced support from friends were most likely to lose weight at 6 months—80.0% did so. This finding is less counterintuitive when one considers that women who lack support in their usual social environment may be precisely those who would seek out and subsequently benefit from supportive group-based classes. A targeted intervention could extend group classes into maintenance, consistent with clinical experience and qualitative work suggesting that participants want ongoing contact (38). Unfortunately, attendance at group classes during weight maintenance is typically low and/or tapers considerably, although individual phone calls with staff during weight maintenance have been very successful (2) as are interaction and feedback via online support groups/websites (e.g., 39). However, given these women report “never” experiencing support from friends, alternatives exist. Rather than influencing weak existing friendships or fostering artificial group relationships, a targeted intervention could encourage women to develop new friendships around healthy behaviors and access new networks that interpersonally and structurally support healthy lifestyle behaviors (rather than weight management per se), such as established walking or hiking clubs. Indeed, developing new friendships with (primarily) normal-weight active individuals is consistent with recommendations informed by network-based interaction models that simulate how the spread of obesity can be slowed “by forcing a shift in cluster boundaries,” i.e., introducing boundaries of normal-weight clusters to overweight/obese persons (35). A second alternative exists. These women actually may have what they need, thus, a targeted intervention could foster greater reliance on self rather than on others. This alternative is consistent with self-determination theory and observational data indicating that individuals with higher autonomous motivation for weight loss lost more weight during a 6-month program than those with lower autonomous motivation (40). Future research could systematically examine a breadth of social support constructs, including social support for lifestyle behaviors, autonomous motivation, and the degree of support received from behavioral group members.

Although the present results may stimulate new ideas for future targeted interventions, and shift away from a ‘one size fits all’ approach, caution is needed given the study’s limitations. The sample was comprised of mostly middle-aged women in structured group-based behavioral weight-loss programs. Thus, further systematic research needs to examine whether the signal detection findings are generalizable to younger women, men of all ages, individuals attempting weight loss on their own, and individuals with comorbidities given these samples may experience different levels of initial support and sabotage than women in structured group-based behavioral weight-loss programs. In addition, the sample sizes for the individual ethnic minority groups were small, underscoring the need for future research to examine the role of social support among different ethnic minority groups, such as among Asian Americans or multi-ethnic/racial individuals, given these minority groups in turn represent heterogeneous groups. Other study limitations include those discussed above such as the state of the sabotage subscales and the reliance on self-reported perceptions of support. Despite these limitations, the study’s strengths included systematically expanding the scales’ psychometric work from cross-sectional samples of young adults to middle-aged overweight and obese women from diverse ethnic/racial and educational backgrounds who were actively participating in group-based behavioral weight-loss programs. In addition, the study’s strengths included the predictive use of signal detection methods within a weight-loss sample that identified multiple subgroups with different characteristics, notably differentiating between friend and family support. In summary, the current study represents a useful step towards advancing innovative and methodologically rigorous research regarding the social context of weight loss.