Body Image 12 (2015) 44–52

Contents lists available at ScienceDirect

Body Image journal homepage: www.elsevier.com/locate/bodyimage

The Family Fat Talk Questionnaire: Development and psychometric properties of a measure of fat talk behaviors within the family context Danielle E. MacDonald a,b,∗ , Gina Dimitropoulos b,c , Sarah Royal a,b , Andrea Polanco a , Michelle M. Dionne a a

Ryerson University, Toronto, Ontario, Canada University Health Network, Toronto, Ontario, Canada c University of Toronto, Toronto, Ontario, Canada b

a r t i c l e

i n f o

a b s t r a c t

Article history: Received 26 March 2014 Received in revised form 3 October 2014 Accepted 5 October 2014 Keywords: Fat talk Body image Family Psychometrics Exploratory factor analysis Confirmatory factor analysis

Fat talk has been well studied in female peer groups, and evidence suggests it may also be important in family contexts. However, no instrument exists to validly assess fat talk within the family. The purpose of this study was to develop a measure of fat talk within families and to establish its psychometric properties in young adult women. In Study 1, the Family Fat Talk Questionnaire (FFTQ) was developed and exploratory factor analysis suggested a 2-factor structure (“Self” and “Family” fat talk), and strong internal consistency. Study 2 confirmed its 2-factor structure using confirmatory factor analysis. Study 3 demonstrated the construct validity of FFTQ scores, including significant correlations with related constructs and predictable gender differences. Study 4 demonstrated the stability of FFTQ scores over two weeks. Therefore, the FFTQ produces valid and reliable scores of fat talk behaviors both exhibited and observed by young adult women within the family context. © 2014 Elsevier Ltd. All rights reserved.

Introduction Sociocultural messages about beauty often permeate social interactions and patterns of communicating (Smolak & Levine, 2001). The term fat talk was coined to describe negative bodyrelated conversations that occur between female adolescents (Nichter & Vuckovic, 1994). More specifically, fat talk has been defined as a normative, back-and-forth conversation pattern in which one or more girls/women makes disparaging comments about her own body (e.g., “I’m so fat!”), which leads the other girls/women involved to either negate the comments (e.g., “No you’re not!”) or to similarly disparage themselves (“No, I’m so fat!”; Nichter, 2000). Fat talk appears to serve numerous functions including the management of interpersonal relationships, the strengthening of emotional connections to peers, eliciting reassurance about one’s weight, preventing peer rejection (Nichter, 2000; Nichter & Vuckovic, 1994), as well as facilitating upward and downward social comparisons within female peer groups (Bailey & Ricciardelli, 2010). Males also engage in fat talk; however, the content of their conversations differs from women’s conversa-

∗ Corresponding author at: Department of Psychology, Ryerson University, 350 Victoria Street, Toronto, Ontario, Canada M5B 2K3. Tel.: +1 416 458 5840. E-mail address: [email protected] (D.E. MacDonald). http://dx.doi.org/10.1016/j.bodyim.2014.10.001 1740-1445/© 2014 Elsevier Ltd. All rights reserved.

tions (Engeln, Sladek, & Waldron, 2013). Fat talk is correlated with body dissatisfaction in both adolescent girls and women (Sharpe, Naumann, Treasure, & Schmidt, 2013), and body dissatisfaction increases immediately following experimental exposure to fat talk (Stice, Maxfield, & Wells, 2003). This latter finding suggests a temporal relationship between fat talk and state body dissatisfaction. Research has also shown positive correlations between fat talk, body shame, and restrained eating (MacDonald Clarke, Murnen, & Smolak, 2010; Royal, MacDonald, & Dionne, 2013). Although most research has focused on fat talk within peer groups, fat talk may also occur in, and have important implications within, the family context. Parental overvaluation of appearance and achievement of a low body weight may contribute to body dissatisfaction and under- or overeating in children (e.g., Keery, Boutelle, van den Berg, & Thompson, 2005; Kluck, 2008, 2010). Additionally, mothers who discuss weight may be more likely to have daughters with disordered eating (Fulkerson, McGuire, Neumark-Sztainer, Story, French, & Perry, 2002; Keery et al., 2005; Neumark-Sztainer, Bauer, Friend, Hannan, Story, & Berge, 2010). Furthermore, negative comments about appearance and appearance-related teasing by both parents and siblings is related to weight reducing practices, body dissatisfaction, low self-esteem, depression, and disordered eating in adolescent girls and young women (Eisenberg, Berge, Fulkerson, & Neumark-Sztainer, 2012; Keery et al., 2005; Kluck, 2010).

D.E. MacDonald et al. / Body Image 12 (2015) 44–52

Despite these findings, there is no published measure that adequately assesses fat talk in the family context. There are three validated measures of fat talk in peer contexts (i.e., EngelnMaddox, Salk, & Miller, 2012; MacDonald Clarke et al., 2010; Royal et al., 2013), but these measures do not query about fat talk within the family. Although the Parental Influence Questionnaire (Abraczinskas, Fisak, & Barnes, 2012) and Caregiver Eating Messages Scale (Kroon Van Diest & Tylka, 2010) assess parental influence on body image and eating behaviors, neither assesses family fat talk. Neither of these measures focuses on the body parts that are the specific targets of fat talk discussion, and the latter focuses primarily on eating-related messages. Furthermore, research on peer fat talk shows that both sides of the fat talk conversation are important (Salk & Engeln-Maddox, 2011), but neither measure assesses the respondent’s behaviors. Given the described relationships between negative comments and teasing about appearance from family members and elements of psychological wellness such as body dissatisfaction, restrained eating, and eating disorder symptoms, a psychometrically sound measure of fat talk that is specific to the family context is needed in this area. The Current Study Accordingly, the first goal of this study was to develop a measure of family fat talk by adapting a psychometrically sound measure of peer fat talk for undergraduate women – the Fat Talk Questionnaire (FTQ; Royal et al., 2013) – to be appropriate for use within the family. The second goal was to establish the family version of the FTQ’s preliminary psychometric properties in young adult women (35 and younger), including its factor structure, internal consistency, construct validity, and temporal stability. We chose to focus only on young women for this preliminary psychometric investigation (a) to be consistent with previous research on fat talk, (b) because research has indicated that the nature of fat talk differs by gender, and (c) because we expected that fat talk specifically within the family context might differ between older and younger women given their different roles within the family. Ethics approval was obtained from the university Research Ethics Board for all four studies reported within this paper. Study 1: Development and Exploratory Factor Analysis The goal of Study 1 was to develop the Family Fat Talk Questionnaire (FFTQ) items and to examine its internal factor structure using exploratory factor analysis among young adult women. Method Questionnaire development. The FFTQ items were developed by adapting the FTQ items (Royal et al., 2013) to be appropriate for use within the family. The FTQ is a 14-item self-report scale that asks respondents to indicate the frequency with which they engage in various fat talk behaviors when they are with similarweight female peers (e.g., “When I am with one or several close female friends, I complain that my stomach is fat”). Items are rated on a 5-point scale ranging from Never to Always. FTQ scores are computed by summing the responses for the 14 items. The FTQ consists of a single factor, and the items were found to be internally consistent (˛ = .94) and temporally stable (r = .90) over two weeks in undergraduates (Royal et al., 2013). Construct validity in undergraduates has been shown using significant correlations with measures of body dissatisfaction, body shame, body surveillance, restrained eating, social physique anxiety, and peer fat talk; a nonsignificant correlation with socially desirable responding; and significant differences in frequency of fat talk between men and women (Royal et al., 2013).

45

The first three authors developed the FFTQ items. The first and third authors were female PhD candidates in clinical psychology who specialize in eating disorders and have previous experience with scale development in the area of fat talk. The second author is a female PhD-level social worker who specializes in family issues related to eating disorders. We chose to adapt the FTQ items rather than develop new items because the original FTQ was originally developed using a rigorous qualitative methodology, and its content was reduced from a comprehensive list of 62 items. Additionally, the FTQ’s psychometric properties in undergraduate women were rigorously investigated and very strong. The rigorous item development is a strength of the FTQ, and as such, we chose to adapt these items as we expected that the content of typical fat talk comments would be similar between peer and family contexts. Additionally, by adapting the FTQ, there is correspondence between the peer and family fat talk constructs as assessed by these measures. The 14 items of the FTQ were adapted in two ways to assess both sides of fat talk conversations within the family: First, to assess the respondent’s own behaviors when interacting with her family members in the past year; and second, to assess behaviors that the respondent observed her family members engaging in during the past year. For example, the FTQ item “When I’m with one or several close female friend(s), I complain that I am fat” was adapted to include both of the following: “When I’m with my family members, I complain that I am fat,” as well as “When I’m with my family, I hear them complain that they are fat.” We elected to adapt the items in both ways because of the social nature of fat talk (Nichter, 2000) and because both sides of the conversation are important (Salk & Engeln-Maddox, 2011), particularly in the family context. Items were preceded by these instructions: “We are interested in the comments you say out loud when you are with your family members over the last year. We are also interested in the comments your family members made about their bodies over the last year. We define family broadly to include parents, siblings, partners, etc. Please keep this in mind when filling out the following questions.” We added this one-year timeframe (which the FTQ does not use) to ensure that participants reflected upon current behaviors, as it is possible that family fat talk may change as individuals pass through different developmental phases and as relationships with family members evolve over time. Items were rated on a 5-point scale from 1 (Never) to 5 (Always). After FFTQ items were developed, we independently consulted with two experts who are clinical psychologists specializing in eating disorders and have experience in body image scale development—neither was involved with the present study. Both experts agreed that the family fat talk construct had been comprehensively surveyed in the FFTQ items. Participants. Participants were female undergraduates (N = 278) who were recruited from the undergraduate psychology research participant pool. Ages ranged from 17 to 35 years (M = 19.1, SD = 2.6). The sample was ethnically diverse. The most common ethnicities represented were Caucasian (45.5%), East Asian (12.3%), South Asian (10.8%), mixed ethnicity (8.3%), Southeast Asian (7.9%), and Black (6.9%). Participants had a mean body mass index (BMI) of 22.2 kg/m2 (SD = 4.2). Measures. Preliminary Family Fat Talk Questionnaire (FFTQ). The 28item preliminary FFTQ, described above, was given to participants. Demographics questionnaire. Basic demographic information was collected, including age, gender (to confirm that they were female), ethnicity, and self-reported height and weight. Body mass

46

D.E. MacDonald et al. / Body Image 12 (2015) 44–52

Table 1 Factor loadings for exploratory factor analysis. Item

Factor 1

2

1. When I’m with my family members, I complain that my arms are too flabby. 2. When I’m with my family members, I complain that my body is out of proportion. 3. When I’m with my family, I complain that I am fat. 4. When I’m with my family, I complain that I should not be eating fattening foods. 5. When I’m with my family, I complain that my clothes are too tight. 6. I criticize my body compared to my family members’ bodies. 7. When I’m with my family members, I complain that I feel pressure to be thin. 8. When I’m with my family members, I complain that I’m not in shape. 9. When I’m with my family members, I hear them complain that their arms are too flabby. 10. When I’m with my family, I hear them complain about the proportion of their bodies. 11. When I’m with my family, I hear them complain that they are fat. 12. When I’m with my family, I hear them complaining that they should not be eating fattening foods. 13. When I’m with my family, I hear others complain that their clothes are too tight. 14. When I’m with my family members, I hear them criticize their bodies compared to their family members’ bodies. 15. When I’m with my family members, I hear them pressure each other to be thin. 16. When I’m with my family members, I hear others complain that they are not in shape.

−.10 .06 −.13 .00 .04 .09 .11 .06 .66 .79 .79 .68 .68 .73 .59 .72

.68 .63 .95 .61 .66 .64 .63 .68 −.02 .00 .01 −.06 .06 .02 .11 −.05

Note: Bolded values indicate significant factor loadings. N = 278.

index [BMI = weight (kg)/height (m2 )] was calculated for each participant. Procedure. The study was conducted online using Qualtrics software. Informed consent was obtained electronically prior to completing the questionnaires, and an electronic debriefing form appeared on the computer after the questionnaires explaining the purpose of the study. Participants were awarded partial course credit (1%) as compensation. Results Item analysis and reduction. In terms of data cleaning and missing data, 0.5% of total item-level data were missing. Data appeared to be missing at random, with no participants systematically missing multiple item responses, and no items systematically missing responses from numerous participants. Missing values were estimated using the mean of the other items focusing on the same target (i.e., respondent or family member; Tabachnick & Fidell, 2001). Initial item analysis was conducted on the 28 items and did not reveal any items with low item-total correlations (r < .30). Initial Cronbach’s alpha was very high, ˛ = .95, indicating redundancy between the items. In terms of item reduction, it was decided a priori that it was most desirable to conduct the item deletion such that both items from pairs with matching content (i.e., respondent’s behavior and family’s behavior) should be either retained or deleted, in order to preserve consistency in the reduced scale. The item reduction strategy involved examining inter-item correlations, and deleting one item from each pair with high or moderately high inter-item correlations. This was done in a stepwise manner such that the two items with the highest inter-item correlation were examined with respect to item content, and one item was selected for deletion based on a face-valid assessment of which of the two items appeared to be a better overall fit with the construct. This process resulted in the deletion of 12 items, leaving a total of 16 items. Cronbach’s alpha was .90 for these 16 items. Exploratory factor analysis. An exploratory factor analysis (EFA) using principal axis factoring was conducted on the retained 16 items, using an oblique rotation (promax), because we expected any emerging factors to be correlated. The Kaiser–Meyer–Olkin measure revealed appropriate sampling adequacy, KMO = .92. Barlett’s test of sphericity, 2 (120) = 2124.78, p < .001, indicated that the inter-item correlations were sufficiently large to conduct an

EFA. Parallel analysis was utilized to determine the eigenvalue cutoff for retaining factors, as this is recommended as being a more rigorous threshold than the commonly used Kaiser’s eigenvalue criterion of 1 (Fabrigar, Wegener, MacCallum, & Strahan, 1999). O’Connor’s (2000) SPSS syntax for parallel analysis was utilized, which runs 1000 iterations of random data eigenvalues in order to set the eigenvalue threshold for the 95th percentile of a random sampling distribution, based on the present sample characteristics (i.e., Cases = 278, Variables = 16). Two components emerged from the EFA, and both of their eigenvalues exceeded the corresponding eigenvalues produced by the parallel analysis (Criterion 1: 1.52; Criterion 2: 1.40). Combined, the two extracted factors explained 49.8% of the variance (Factor 1: eigenvalue = 6.11, 38.22% variance explained; Factor 2: eigenvalue = 1.85, 11.55% variance explained). The significance level for factor loadings was set at r = .46, p < .001, which was determined by doubling the critical value for a Pearson correlation using the selected alpha level and sample size of N = 278 (Stevens, 2002; a very conservative p-value was selected due to the large sample size). The pattern matrix was examined and revealed that the first eight items loaded onto Factor 2, and the latter eight items loaded onto Factor 1. No items loaded onto both factors. See Table 1 for item-factor loadings. In order to further validate the fit of this solution, the EFA was recomputed with a forced 2-factor solution. As expected, the results of the forced solution (eigenvalues, variance explained, and factor loadings) were completely identical to the unconstrained EFA. The results of Study 1 therefore indicate that the FFTQ has two factors, each with eight items. The item content clearly distinguished the factors, with Factor 1 reflecting fat talk exhibited by the respondent’s family members (“Family” subscale) and Factor 2 reflecting the respondent’s behavior (“Self” subscale). Cronbach’s alphas and subscale scores (comprised of item totals) were computed for the applicable items (Family: ˛ = .89, M = 19.9, SD = 6.6; Self: ˛ = .88, M = 17.9, SD = 6.6). Study 2: Confirmatory Factor Analysis The goal of Study 2 was to confirm the two factor structure of the FFTQ obtained in Study 1 using confirmatory factor analysis. Method Participants. Participants were 174 women recruited online using Cognilab/Mechanical Turk. Amazon Mechanical Turk is a crowdsourcing online labor market which allows recruitment of

D.E. MacDonald et al. / Body Image 12 (2015) 44–52

individuals (“workers”) to complete “microtasks” for a nominal payment, determined by the task “requester” (see Mason & Suri, 2012 for a detailed description of this platform). Mechanical Turk is only available to requesters in the United States (though the “workers” are sourced globally), but the Cognilab system is available to Canadian researchers and is directly integrated to recruit participants from Mechanical Turk’s participant pool. Research recruitment using Mechanical Turk is becoming increasingly popular, and recent evaluations of this method indicate that high quality data can be obtained using this platform quickly and with minimal expense (Buhrmester, Kwang, & Gosling, 2011; Mason & Suri, 2012; Paolacci & Chandler, 2014). Further, reliable and valid data on body image have been gathered via Mechanical Turk (Gardner, Brown, & Boice, 2012). Participants were women ranging in age from 19 to 35 (M = 27.9, SD = 4.0). The mean BMI (computed from self-reported height and weight) was 27.9 (SD = 7.6). The sample was ethnically diverse, with the most common reported ethnicities being South Asian (50.3%), Caucasian (30.1%), Southeast Asian (6.4%), and Black (3.5%), with the remainder of the sample comprising various other ethnicities. Procedure. Study 2 was advertised on Cognilab/Mechanical Turk as “Psychology research study related to body image and social behaviors (females only, approx. 5 mins).” If interested, participants could click on the task, which brought them to a consent form. Participants were paid $0.25 for their participation. After consent, they were taken to the task, which consisted of the 16-item FFTQ derived in Study 1 (Cronbach’s alpha in the present sample: Self, ˛ = .90; Family, ˛ = .93), followed by the demographic questionnaire. Gender was included in the demographic questionnaire in order to screen out male respondents. Results The confirmatory factor analysis was conducted using MPlus version 7 (Muthén & Muthén, 2012). Missing data occurred in 0.0% of the FFTQ items, though a small number of participants failed to report demographic variables (e.g., ethnicity), which were kept as missing. Individuals who failed to report age and/or gender were excluded. Based on the results of the EFA, a 2-factor solution was specified, with one latent factor for the Self subscale and one latent factor for the Family subscale. Because the distributions for some of the items were slightly skewed, we used Maximum Likelihood with Robust Estimation (MLR), which adjusts model fit indices to accommodate non-normality (Muthén & Muthén, 2012). Model fit was considered good if the root-mean square error of approximation (RMSEA) < .06, and both the comparative fit index (CFI) and the Tucker-Lewis index (TLI) > .95 (Hu & Bentler, 1999). In addition, because the subscales consist of item pairs matched in content and which differed only slightly with respect to the individual referred to by the item (i.e., respondent or family member), we estimated error correlations for item pairs across factors. For example, “When I’m with my family, I complain that my clothes are too tight,” which loads on the Self subscale, is a mirror image of the item “When I’m with my family, I hear others complain that their clothes are too tight,” which loads onto the Family subscale. When the wording of pairs of items are highly similar, it is often the case that errors for these items are highly correlated, reflecting measurement error that is due to a shared method effect (Brown, 2003; Kenny & Kashy, 1992). Thus, we planned a priori to freely estimate the cross-factor error covariances for the eight pairs of items. Because robust estimation was used, we conducted a scaled chi-square difference test to compare model fit (Satorra, 2000). The results showed that the two factor model including error covariances among the eight pairs of items provided a good fit to the data, scaled 2 (95) = 152.58, p < .001, RMSEA = .06, CFI = .96,

47

Table 2 Factor loadings for confirmatory factor analysis. Item

Standardized factor loadings Factor 1 (Family)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Factor 2 (Self)

R2

.63 .73 .81 .75 .77 .67 .74 .76

.39 .53 .65 .57 .60 .46 .54 .58 .49 .61 .73 .60 .68 .60 .60 .58

.70 .78 .85 .78 .83 .78 .77 .76

Note: Bolded values indicate significant factor loadings. R2 reflects variance explained in each observed item. N = 174.

Table 3 Correlations between latent factors derived by confirmatory factor analysis.

Self factor with family factor Item 1 with Item 9 Item 2 with Item 10 Item 3 with Item 11 Item 4 with Item 12 Item 5 with Item 13 Item 6 with Item 14 Item 7 with Item 15 Item 8 with Item 16

Correlation

Standard error

p

.63 .38 .09 .005 .44 .26 .19 .33 .38

.08 .08 .10 .11 .10 .09 .11 .10 .10

The Family Fat Talk Questionnaire: development and psychometric properties of a measure of fat talk behaviors within the family context.

Fat talk has been well studied in female peer groups, and evidence suggests it may also be important in family contexts. However, no instrument exists...
418KB Sizes 4 Downloads 8 Views