Factor structure and psychometric properties of a perceived barriers to physical activity scale for low-income teens

Background: Despite known benets of health and well-being, high-school aged adolescents don’t engage in regular daily physical activity (PA). PA levels are lesser among low-income adolescents. One signicant predictor of PA levels is perceived barriers, however, there is a lack of a valid and reliable scale to measure perceived barriers among the low-income adolescent population. The objective of this study was to develop a scale to assess perceived barriers to PA among these high-school aged, low-income adolescents. Methods: The Perceived Barriers to PA Survey for Low-Income Adolescents (PBPA-A) was developed using a mixed-method approach. Initially, 110 items were identied from pre-existing surveys and extensive literature review. These items were then tested using cognitive interviews (n=15) to ensure the wording clarity and appropriateness. Items were deleted and revised based on the feedback from the cognitive interviews. Then, a 2-week test-retest reliability was conducted with 105 adolescents to assess the scale’s consistency over time using intraclass correlation coecient (ICC). Items with ICC ≤ 0.4, indicating insucient reliability, were deleted, as were those demonstrating a oor effect. Exploratory factor analysis (n=999) and conrmatory factor analysis (n=999) were used to explore and conrm the underlying structure of the remaining 73-item scale. Items with a loading of ≥ 0.4 were retained and internal consistencies for the whole scale and subscales were calculated using Cronbach’s alpha. Test-retest reliability was recalculated, and concurrent validity was tested in a sample of 1,998 adolescents against the pre-established physical activity questionnaire for adolescent (PAQ-A) using Spearman correlations. Results: The EFA yielded a 7-factor solution with 37 items, which was further cross-validated by CFA. The test-retest reliability was 0.75 for the whole scale and ranged from 0.65 to 0.81 for the subscales. As predicted, the PABA-A scale and subscales were negatively associated the PAQ-A score, indicating adolescents with higher perceived barriers had lower physical activity levels. Conclusions: The PBPA-A is a valid and reliable scale that can be used to


Introduction
The bene ts of adolescents' engagement in physical activity (PA) range from reduced risk of type 2 diabetes and obesity to improved mental health and more favorable sleep patterns [1][2][3][4][5]. Despite these bene ts, adolescents' PA levels have been found to decline with age by approximately 38-41 minutes per year [6,7]. The U.S. Youth Risk Behavior Surveillance Survey 2017 results revealed that only 26.1% of high school adolescents met the U.S. Physical Activity Guidelines of engaging in moderate to vigorous PA for ≥ 60 minutes per day [8,9].
One predictor of low PA levels is perceived barriers. In fact, perceived barriers (one's judgments regarding personal, social, environmental, and economic obstacles that hinder engagement in health behaviors, like PA) have been identi ed as the "most powerful" Health Belief Model's constructs [10]. Yet, studies examining the relationship between adolescents' perceived barriers and PA have been inconclusive [11][12][13][14]. This may be due to the lack of a comprehensive, validated scale to assess them. Some studies examining the relationship between high school-aged adolescents' perceived barriers and PA levels have used scales developed for adults, sometimes with minor adaptations [15,16]. Other studies have developed their own scales, with limited reliability and validity assessment [17][18][19][20]. To better understand the relationship between adolescents' perceived barriers and PA, the use of a comprehensive, reliable, and valid barriers scales is critical.
Notably, PA levels are lower among adolescents of low socioeconomic status (SES). Because low-income adolescents may have different perceived barriers than adolescents from higher SES backgrounds [21] and older adolescents demonstrate lower PA levels, this study sought to develop a scale to assess the perceived barriers to PA among low-income high school-aged adolescents.

Methods
A mixed-method approach was used to develop this Perceived Barriers to PA Survey for Low-Income Adolescents (PBPA-A). Multiple steps, detailed below, were conducted in in the PBPA-A's development, including preliminary work to identify PBPA-A items and conduct item cognitive testing, and eld-testing to assess its reliability, factor (subscale) structures, and the scale's internal consistency and concurrent validity. All statistical analyses were conducted using SAS (version 9.4, SAS Institute, Cary, NC). This work was approved by Rutgers Institutional Review Board (protocol E13-022).

Preliminary Work
An initial list of 77 barriers to PA perceived by adults was obtained from previous research (i.e., from literature review and interviews) conducted by the study team. Some PA barriers identi ed were inclement weather, competing interests, lack of motivation, health problems, time, transportation, cost issues, and lack of knowledge. Another extensive literature review was conducted to identify adolescents' barriers to PA. An additional 33 items were identi ed, including family support and embarrassment. The compiled barrier items (total = 110) were reviewed by the research team and New Jersey Expanded Food and Nutrition Education Program professional staff to establish content validity. Items that were unclearly worded were modi ed. Items were rated using a 5-point Likert scale regarding how often each barrier stopped teens from exercising.
Cognitive interviews were conducted with 15 low-income high school students to further assess the survey's face validity. The adolescents were asked to read the instructions and each item aloud, describe in their own words what the question was asking, and then explain how they would respond using the response choices provided. All cognitive interviews were audio-recorded; no identi able information was collected. Items and responses were assessed for readability, comprehension, completeness, and clarity. Items that were unclear, too complex, did not apply (as written), or had multiple meanings were modi ed based on the interviews' results. Three items were removed due to redundancy. Other changes made were minor; thus, 107 items remained.

Study Population and Recruitment
The eld-test sample was recruited from 12 low-income high schools in which more than 50% of the adolescents quali ed for free or reduced-priced lunch from the U.S. National School Lunch Program and 6 community agencies in low-income areas of New Jersey. Eligible participants were adolescents (grades 9 to 12) who could understand, read, and speak English. Surveys were collected between May 2012 and April 2016 (N = 2,762).

Test-Retest Reliability
A subsample of older adolescents (n = 105) from 5 sites participated in the test-retest assessment. The teens were asked to complete the PBPA-A survey twice, 2 weeks apart. At baseline, their demographic information, such as age, gender, and ethnicity, was provided. Because the questionnaire was lengthy, three versions were created, wherein the order of the barrier items was randomized and 3 attention-check questions (i.e., "On this line, put two checks in the 'Often' column") were placed in different locations such that surveys belonging to adolescents whose attention waned could be removed from the sample. The survey took approximately 10-15 minutes to complete. No compensation was provided. All responses were anonymous and matched by unique identi ers.
Reliability over time was assessed as an early item-reduction step, and again for the revised full PBPA-A and its subscales after factor analysis. Both times this was accomplished by comparing the adolescents' PBPA-A testretest responses for each item on PBPA-A scale. Intra-class correlation coe cients (ICC) were used to compare the responses [22]. ICCs were classi ed as follows: "excellent" (≥ .81), "good" (0.61-0.80), "moderate" (0.41-0.60), or "poor" (≤ 0.40) [22][23][24]. Questions were considered to have poor reliability over time if the ICC was < 0.40 [22].

Preliminary Item-Reduction Steps
Prior to the factor analysis, the survey items were removed if they exhibited: (1) poor test-retest reliability (i.e., ICC < 0.40), or (2) a oor effect (i.e., > 80% of participants rated the item as a "1" ("Never a barrier"). Remaining items were used for exploratory factor analysis and internal consistency testing.

Factor Analysis and Internal Consistency
In order to cross-validate the proposed measure, the sample was randomly split into halves. One half (n = 999) was used to perform an EFA to explore the underlying structure of the survey items, and the other half was used to perform a CFA (n = 999) to con rm the identi ed structure. The samples did not differ regarding gender (χ 2 = 0.10, df = 998, p = 0.74), race/ethnicity (χ 2 = 3.51, p = 0.48), and grades (χ 2 = 2.79, df = 998, p = 0.42).
Prior to the EFA, the items' factorability was examined using the intercorrelation matrix from Pearson's correlations. Items with correlation coe cient (r) values below 0.20 were to be deemed inappropriate for factor analysis [25,26]. Items with r values above 0.80 suggested multicollinearity (i.e., items can be predicted by each other) and were to be eliminated before the EFA [25,26].
The Kaiser-Meyer-Olkin (KMO) test was used to examine sampling adequacy. The KMO test values ranged between 0 and 1. A value of 0 means that the sum of partial correlations is larger relative to the sum of correlations, indicating diffusion in the correlation pattern (hence, again, factor analysis is likely to be inappropriate). The values for KMO tests should be above 0.60 [25,27]. The Bartlett's test of sphericity either con rms or rejects the null hypothesis (that there are no relationships observed between items); thus if relationships are ample to cause the test results to be signi cant (p < 0.05) to reject the null hypothesis and indicate the items are factorable [26].
An EFA using maximum likelihood estimates with oblique rotation (Promax) was conducted to identify the factor structure. The oblique rotation enables the factors' correlations to be examined, so the model can be better interpreted [27,28]. While determining the number of factors, it has been recommended that researchers should utilize different methods [26]. Thus, to determine the best factor solution these models were used: (1) Kaiser-Guttman criteria (i.e., eigenvalue > 1); (2) the scree plot; (3) parallel analysis; (4) minimum average partial (MAP) test; and (5) the interpretability of the factor solution (i.e., whether the items loaded on the same factor had the same concept, the factor loading was following a simple structure solution). Parallel analysis and MAP tests were employed because they were considered superior estimates of the number of factors to retain, based on simulation studies [29,30]. The code used for these test was provided by O'Connor and colleagues [31].
After determining the number of factors, only items that achieved a factor loading > 0.4 and that loaded on only one factor were retained. Internal consistency for the entire PBPA-A an8d for each subscale was measured using the Cronbach's alpha, with acceptable alpha values ≥ 0.70.
Again, to con rm the EFA factor structure identi ed, a CFA was conducted with the second sub-sample.
Although a chi-square test (χ 2 ) is traditionally used to test goodness of t, with statistically signi cant results suggesting that a model exhibits a lack of t to the data, χ 2 results are not always dependable with larger samples so both χ 2 and other t indices were examined [28]. Those used were: the model, the Standardized Root Meant Square Residual (SRMR), the goodness-of-t index (GFI), the Root Mean Square Error of Approximation (RMSEA), the Comparative Fit Index (CFI), and the Bentler-Bonett Non-normed Fit Index (NFI). The model t was assessed using these criteria: GFI, NFI, and CFI values of > 0.90 were considered to be acceptable; SRMR and RMSEA values ≤ 0.055 were considered to be acceptable [28,[32][33][34].
After assessing the model t, the signi cance of the parameter estimates for each item was examined to see if they were different from zero. The parameter estimate is equivalent to a path coe cient from a survey's latent factor. A nonsigni cant parameter estimate suggests the survey item did not signi cantly contribute to the underlying model and should be deleted.

Concurrent Validity
In the absence of a standard for measuring perceived barriers, concurrent validity was assessed by examining the relationship between adolescents' perceived barriers as measured by the PBPA-A and their self-reported PA levels evaluated using the Physical Activity Questionnaire for Adolescents (PAQ-A). The PAQ-A has been deemed the most suitable, valid, and reliable self-report tool for examining adolescents' PA levels during the school year [35][36][37]. PAQ-A items are assessed using a 5-point Likert scale to estimate total activity via items that address the following: spare time PA; PA during physical education classes, lunch, right after school, evening, weekends; and PA and PA frequency for each day during the past week [38]. It was hypothesized that perceived barriers would be negatively associated with PAQ-A scores. Skewness, kurtosis, and the Shapiro-Wilk test for normality were used to assess variables' distributions. None were found to be normally distributed, so the Spearman correlation coe cient was used.

Participants' Characteristics
From among the teens surveyed (N = 2,762), only data those students who responded correctly to the bogus questions (i.e., who were deemed to have given adequate attention to the survey), 1,998 (see Table 1) were included in the analyses.  [39]. Also, measures of sampling adequacy for each item were all above 0.89, (i.e., larger than the recommended value of > 0.70 [39]. Therefore, no additional items were removed prior to the EFA. Four methods were used to determine the number of factors that should be retained. The Kaiser-Guttman criteria (eigenvalues ≥ 1.0), suggested 16 factors, which accounted for 99.9% of the total variance; the "elbow" of the scree plot indicated 7 factors; and the parallel analysis and MAP test suggested 11 and 8 factors, respectively. Thus, separate EFAs were run for the 16-, 11-, 8-, and 7-factor solutions. As per recommendations, the interpretability of the factor solution (i.e., whether the items loaded on the same factor shared conceptual similarity, or the factor loadings followed a simple structure solution) was used to determine the nal factor structure [28]. An attempt was made to include at least 3 items per domain to ensure minimum coverage of each construct's theoretical domain. The 7-factor solution was deemed to be the most interpretable model.
Items that did not load ≥ 0.4 on any factors (n = 28) were deleted, and EFAs were repeated with the remaining items until a simple structure (i.e., no items loading on multiple factors) was obtained. This resulted in 7 factors and 39 items; however, two items -"It takes too much time" and "It does not feel good" -were deleted as they did not t into the factor on which they had loaded (i.e., Factor III). The EFA was rerun and yielded 7 factors of 37 items (Table 2). Although "My friends don't do it" loaded at < 0.4, it was kept due to its conceptual importance.
The nal factor structure is shown in Table 2. The intercorrelations between factors ranged from 0.30 to 0.59.

Internal Consistency and Test-Retest Reliability
The Cronbach's alpha for the PBPA-A scale was 0.94 (including the item "My friends don't do it"), and subscales' alphas ranged from 0.71 to 0.88 (Table 2). Also, the Cronbach's alpha analysis suggested no improvement if any items were removed. The ICC for the full scale was 0.75. Table 2 shows the ICC values for each item and each subscale, all of which were above 0.6.

Concurrent Validity
Because the factor structure was con rmed and the internal consistency and test-retest reliability were reasonable, the adolescents' scores for the PBPA-A and each of its subscales were calculated by computing a mean score for all items loading on them. Table 4 presents the means, standard deviations of the PBPA-A subscales. To establish concurrent validity, the PBPA-A score and the score for each PBPA-A subscale were correlated with the PAQ-A score. As expected, they were negatively correlated (r = − 0.20 to − 0.43; p < 0.0001) ( Table 4).

Discussion
This study addressed a lack of scale development and validation for assessing low income teens' perceived barriers to PA. To the best of the authors' knowledge, this is the rst study that has undergone rigorous testing for factor analysis, reliability, and validity to develop a comprehensive scale for use with this audience.
One contribution of the PBPA-A was the inclusion of barriers regarding preferred engagement in technologyrelated activities (e.g., "I would rather play video games"; "I would rather spend time on the computer"). Preferred involvement in technology-related activities (e.g., television, computer, phone) has been identi ed as a signi cant PA barrier for high school-aged adolescents in qualitative studies [16,40,41]; however, no previously developed scales have addressed these concepts [16,18,19,42,43]. Due to teens' high prevalence of technology-based device usage, the inclusion of these concepts provides a more comprehensive picture to document low-income adolescents' PA barriers.
Contrary to previous PA barrier scales developed for teens [16,18,19,42], two commonly assessed barriers, safety and injury, did not end up in the nal PBPA-A. During the initial development of the PBPA-A, 9 items related to safety and 2 items related to injury were included among the 110 items. However, some items were removed due to oor effect, and none of the others exhibited su cient ICCs or EFA loadings. Further, over 70% of the adolescents responded "Never" or "Once in a while" to all 11 items, suggesting to most they were not perceived barriers. Notably, safety and injury have been found to be rated low by adolescents in other studies [16,19,42]. Although it is counterintuitive that lack of a safe environment or having an injury or medical condition would not be perceived barriers to PA, previous research has suggested that adolescents perceive themselves as invulnerable [44]. Thus, while these factors may indeed be barriers to PA, they are not perceived as such by this population.
Previous research examining the factor structure of adolescents' perceived PA barriers is somewhat limited.
Multiple studies have reported on adolescents' barriers [16-20, 42, 43], yet only ve provided detailed information about the items they included in the scales used [16,18,19,42,43], and only one conducted a factor analysis [16]. That study, conducted by Allison and colleagues, examined the factor structure of a 16-item perceived barrier scale among 1,041 high school students [16]. The results yielded two factors: internal barriers (10 items) and external barriers (6 items). The internal barrier factor included items on individual, physical, or psychological barriers such as discomfort, lack of energy, and self-consciousness. The external barrier factor loaded items regarding lack of social support, facilities, part-time job, and cost. However, both the internal and external barrier scales encompassed single items that are actually multidimensional constructs. From a nutrition education perspective, it would be more useful to use multiple items to assess these constructs to aide in better targeted interventions. For example, "self-consciousness" is a perceived barrier in both Allison et al.'s scale and the PBPA-A. However, in the former scale, self-consciousness is a single item, whereas in the PBPA-A it is a "factor" comprised of 6 items re ecting different constructs, such as "People making fun of me" and "I don't think I can do it." Approaches used and messages tailored to overcoming these barriers would be entirely different depending on the scale's content. For example, to address the former construct professionals may aim to build self-esteem and/or to teach how to deal with bullying, whereas education tailored to the latter would be designed to increase self-e cacy.
Moreover, this study suggests some barriers cannot be classi ed as internal or external, as responses depend on respondent interpretation. For example, a girl's discomfort (identi ed as an internal barrier by Allison and colleagues) is an internal barrier if it is due to lack of coordination or her not feeling she is good at sports.
However, it is an external barrier when it is due to discomfort caused by unfavorable weather. These problems highlight the complexity of studying perceived barriers and the need for a scale that can better assess concepts within broader barrier constructs.
The factor analysis done for the PBPA-A, identi ed 7 theoretically meaningful factors (self-consciousness, competing interests, personal expectations and familial in uences, lack of resources, unfavorable weather, lack of motivation, and competing priorities (i.e., job, chores) with 37 items. With multiple item loadings on each factor, the PBPA-A is able to more thoroughly examine PA barriers, reduce ambiguity, and better examine the complexity of the perceived barriers. The correlations among factors (shown in Table 3), all less than 0.70, suggest that each factor likely measures a unique concept [45]. The PBPA-A' theoretical structure having been cross-validated by CFA strengthens the evidence that this structure is acceptable.
Additionally, this study showed evidence of the PBPA-A's and its subscales' reliability (i.e., the internal consistency and test-retest reliability). These reliability results were comparable to the two previously published scales that reported their Cronbach's alphas, which ranged from 0.73 to 0.90 [16,43], and the one scale that reported a test-retest reliability ICC of 0.90) [43]. These reliability results suggest that the PBPA-A can be used in its entirety or separately with each subscale.
Multiple health behavior models indicate that perceived barriers impact behavioral outcomes, like PA [10,[46][47][48]]. Yet, systematic reviews have yielded mixed results. Some have found perceived barriers had a small to moderately negative association [11,37], while others were inconclusive or found no association [12,13]. These inconsistencies might be caused by lack of the use of a valid and reliable scale, considering all the scales developed previously had limited reliability and validity information [16,18,19,42,43]. The strong negative correlation demonstrated by the concurrent validity results of the PBPA-A indicates a better alignment with the expected impact of the barriers on PA behaviors [10,[46][47][48] Limitations One notable study limitation was the use of a convenience sample in New Jersey, which might limit the ndings' generalizability. Nonetheless, since the study sample was relatively large and racially/ethnically diverse with an approximately equal number of male and female participants, its generalizability to low-income teens throughout the U.S. and other developed countries may be reasonably good. Additional research to con rm the PBPA-A's factor structure, reliability, and validity in other geographical areas would be prudent.
Another limitation was the lack of a "gold standard" to be used for concurrent validity testing. Yet PBPA-A's concurrent validity was established through the notion that those with more barriers are less active, and this was done using a valid and reliable tool (i.e., the PAQ-A). Further validity studies to establish criterion validity against objective measures may be warranted.

Conclusion
This study is one of few studies done to develop a scale to assess perceived barriers to PA among older adolescents. The ndings reported herein provide empirical support for its reliability and validity. Jersey (#E13-022). Informed consent was obtained in writing forms from all participants before data collection.

Consent for publication
Not applicable.
Availability of data and materials The datasets used and/or analyzed may be available from the corresponding author on a reasonable request after IRB approval.