Demographic characteristics of the participating girls are shown in Table
1. Average age of the girls was 15.74 years (SD = 0.77) and the average BMI was 20.91 kg/m
2. Household income was unfairly distributed across the four income categories. Some %62.8 of the participating girls' family had < $320 household income, %48.5 (n = 248) of fathers had completed high school, with %16.8 (n = 86) completing a college or graduate degree, and %54 (n = 276) of mothers had completed high school, with %09.0 (n = 46) completing a college or graduate degree.
Table 1
Characteristics of the participating girls
Age (years) | 15.74 (0.77) |
Weight | 54.22 (11.39) |
Height | 161.51 (10.88) |
BMI | 20.91 |
Father education (%) | |
Some high school | 31.5 |
High school graduate | 48.5 |
College or graduate degree | 16.8 |
Up BM | 3.2 |
Mother education (%) | |
Some high school | 35.6 |
High school graduate | 54.0 |
College or graduate degree | 09.0 |
Up BM | 1.4 |
Household income (%) | |
< $320 | 62.8 |
$321–$550 | 25.8 |
$551–1100 | 9.6 |
> $1100 | 1.8 |
Analysis approach
The sample size of 512 was sufficient to produce reliable correlation coefficients so that popular textbooks on factor analysis give specific advice on sample size for factor analysis. The required variable to subject ratio lies between 1:5 and 1:10 (14, 30).
Prior to performing Principal Components Analysis (PCA), the suitability of data for factor analysis was assessed. An inspection of the correlation matrix in each subscale revealed that most of the correlations were greater than 0.30, therefore, some clustering of items was expected and exploratory factor analysis was deemed appropriate in the early stage of research [
31]. The Kaiser-Meyer- Olkin Measures of Sampling Adequacy value for examined scales ranged from 0.61 to 0.93, exceeding the recommended value of 0.60 [
31] and the Bartlett's test of sphericity [
32] reached statistical significance (P < 0.001), supporting the factorability of the correlation (Table
2). Thus, Principal Component Analysis (PCA) was used to identify scales' dimensions in this study. The decision between orthogonal and oblique rotation was made, examining the correlations among the factors [
31]. The results of factor analysis are presented here:
Table 2
KMO* & Bartlett's test of sphericity psychosocial determinants of physical activity
Self-efficacy | 0.88 | 1043.051 (p = 0.00001) |
Social support | 0.79 | 1181.322 (p = 0.00001) |
Pros & Cons | 0.79 | 1153.842 (p = 0.00001) |
Change strategies | 0.93 | 2243.043 (p = 0.00001) |
Environment | 0.61 | 0387.790 (p = 0.00001) |
For physical activity self-efficacy (Table
3), one factor was identified which was accounted for 55% of the variability in the items. The internal consistency estimate (alpha = 0. 84) was excellent and the test-retest reliability coefficient (r = 0.68) was substantial.
Table 3
Factor analysis for physical activity self-efficacy scale (N = 512)
Be physical activity even it raining or hot | 0.79 |
Get up early, even on weekends, to do physical activity | 0.79 |
Set aside time for physical activity on most days | 0.77 |
Be physical activity even I have a lot of schoolwork | 0.75 |
Be physical activity even my family want me to do something else | 0.71 |
Be physical activity even I feel sad or stress | 0.65 |
Eigen value | 3.35 |
% variance explained | 55.97 |
Choronbach's alpha | 0.84 |
Mean inter item correlation | 0.47 |
Pearson test-retest* | 0.68 |
Two sub-scales were identified for the physical activity social support (Table
4), family support and friend support. These two factors accounted for 55% of the variability in the items. The internal consistency estimate for the family support scale was substantial (alpha = 0.72), as was the internal consistency estimate for the friend support scale (alpha = 0.77). The test-retest reliability of both scales was moderate (r = 0.56 and r = 0.54, respectively).
Table 4
Factor analysis for physical activity social support scale (N = 512)
Friend support
| | |
Tease from your friends | 0.80 | |
Ask from your friends to walk | 0.77 | |
Tell you that you are doing well | 0.73 | |
Do physical activity with you | 0.70 | |
Encourage you to do physical activity | 0.65 | |
Family support
| | |
Done with you | | 0.73 |
Provided transportation | | 0.73 |
Encourage you to do physical activity | | 0.72 |
Watched you | | 0.72 |
Eigen value | 2.62 | 2.32 |
% variance explained | 29.18 | 25.83 |
Choronbach's alpha | 0.77 | 0.72 |
Mean inter item correlation | 0.41 | 0.40 |
Pearson test-retest* | 0.56 | 0.54 |
| Factor1(Factor loadings) | Factor2(Factor loadings) |
For the physical activity decisional balance (Table
5), two sub-scales of pros and cons were also identified, accounting for 50% of the variability in the items. The internal consistency estimate for the pros scale was substantial (alpha = 0.81), as was the internal consistency estimate for the cons scale (alpha = 0.69). The test-retest reliability of both scales was moderate (r = 0.44 and r = 0.36, respectively).
Table 5
Factor analysis for physical activity Pros & Cons scale (N = 512)
Pros scale
| | |
More energy | 0.79 | |
Feel better | 0.76 | |
Help to stay fit | 0.72 | |
Parents would be happy | 0.72 | |
Have fun with my friends | 0.70 | |
Cons scale
| | |
Takes time from being with my friends | | 0.70 |
Too much help from my parents | | 0.65 |
There is too much to learn | | 0.64 |
Don't like physical activity to make me feel | | 0.63 |
Feel embarrassed | | 0.42 |
Eigen value | 3.07 | 1.98 |
% variance explained | 30.70 | 19.88 |
Choronbach's alpha | 0.81 | 0.69 |
Mean inter item correlation | 0.46 | 0.23 |
Pearson test-retest* | 0.44 | 0.36 |
Principal Component Analysis (PCA) with oblique was performed on the students' responses to the 15 change strategies items. Oblique rotation, which allows the factors to be statistically related [
31], was used because it was expected that the factors underlying change strategies would be correlated in reality. An initial analysis with principal component analysis was conducted to identify the number of factors with eigenvalues of 1.0 or greater, which is an estimate of the maximum number of stable factors [
31]. The scree test [
33], suggested the existence of factors.
The eigenvalues for the first 2 consecutive components were 5.70 and 1.06. Examination of the eigenvalues greater than 1 indicated that a 2-factor solution may be appropriate. The examination of the scree plot also suggested that 2 dimensions underlie change strategies scale. Although these two methods are the most popular heuristic, they are potentially unreliable [[
34,
31], and [
35]]. For example, Zwick and Velicer have argued that using eigenvalues greater than 1 to determine the number of factors to extract leads to 'overfactoring', it remains more factors than is optimally required. In this study parallel analysis (PA) [
36] was employed to ascertain the optimal number of factors to extract. The PA requires the researcher to randomly generate a raw data matrix on the same 'rank' as the actual raw data matrix. For example, if one had a 1-to-5 Likert scale data for 512 subjects on 15 variables, a 512-by-15 raw data matrix consisting of 1s, 2s, 3s, 4s and 5s would be generated. These random data can be factor analysed to produce a set of eigenvalues. The eigenvalues associated with the matrix of association based on observed data are also computed. The number of extractable factors is equal to the number with observed eigenvalues greater than the point on the plot where the observed and random eigenvalues cross [[
34,
36], and [
37]].
Using the procedure recommended by Thompson and Daniel [
37], 50 random data sets were generated of the same order of change strategies scale data. The 50 data sets were factored. The mean eigenvalues for the first 8 consecutive components were 1.29 1.22, 1.18, 1.14, 1.10, 1.06, 1.02 and 1.006. Thus, only the first 1 eigenvalues change strategies scale factor analysis exceeded its associated eigenvalues derived from the random data and a 1-factor model was appropriate. This factor is accounted for 38.06% of the variability in the items. The internal consistency estimate for change strategies factors scale was substantial (alpha = 0. 78) and the test-retest reliability was also substantial (r = 0.74) (see Table
6).
Table 6
Factor analysis for physical activity change strategies scale (N = 512)
I say positive things to myself about physical activity | 0.71 |
I set goals to do physical activity | 0.70 |
I do things to make physical activity more enjoyable | 0.70 |
When I get off track with my physical activity plans, I tell myself I can start again and get right back on track | 0.70 |
I keep track of how much physical activity I do | 0.67 |
I reward myself for being physically active | 0.66 |
I look for information about physical activity or sports | 0.61 |
I try different kinds of physical activity so that I have more options to choose from | 0.59 |
I make back-up plans to be sure I get my physical activity | 0.59 |
I put reminders around my home to be physically active | 0.57 |
I find ways to get around the things that get in the way of being physically active | 0.56 |
I have a friend or family member who encourages me to do physical activity | 0.55 |
I try to think more about the benefits of physical activity | 0.53 |
I think about the benefits I will get from being physically active | 0.52 |
I think about how my surroundings affect the amount of physical activity I do (Surroundings are things like having exercise equipment at home or a park near by) | 0.46 |
Eigen value | 5.7 |
% variance explained | 38.6 |
Choronbach's alpha | 0.78 |
Mean inter item correlation | 0.34 |
Pearson test-retest* | 0.74 |
For the physical activity environmental factors, two sub-scales were also identified. The eigenvalues for the first 2 consecutive components were 2.03 and 1.009. Examination of the eigenvalues greater than 1 indicated that a 2-factor solution may be appropriate. The examination of the scree plot also suggested that 2 dimensions underlie environmental factors scale. In this study parallel analysis (PA) [
36] was also employed to ascertain the optimal number of factors to extract. The mean eigenvalues for the first 3 consecutive components were 1.09 1.028 and 1.003. Thus, only the first 1 eigenvalues environmental factors scale factor analysis exceeded its associated eigenvalues derived from the random data and a 1-factor model was appropriate. This factor is accounted for 50.87% of the variability in the items. The internal consistency estimate for environmental factors scale was substantial (alpha = 67) and the test-retest reliability was moderate (r = 0.38) (see Table
7).
Table 7
Factor analysis for physical activity environmental factors scale (N = 512)
It is safe to walk | 0.78 |
Can get to easily | 0.73 |
Enough supplies at Home | 0.72 |
It is difficult to walk | 0.60 |
Eigen value | 2.03 |
% variance explained | 50.87 |
Choronbach's alpha | 0.67 |
Mean inter item correlation | 0.25 |
Pearson test-retest* | 0.38 |
The findings showed intercorrelations among the physical activity-related psychosocial measures. Physical activity self-efficacy was significantly and positively correlated with the physical activity pros scale (perceived benefits) and change strategies, while it was negatively correlated with the physical activity cons scale (perceived barrier).Those girls with higher scores on physical activity self-efficacy reported higher scores on physical activity pros, change strategies and lower scores on the physical activity cons.
Reliability
Reliability was determined by examining both the internal consistency and test-retest stability of the physical activity-related psychosocial measures. The physical activity-related psychosocial measures showed adequate internal consistency (i.e., > 0.70) [
27] with the exception of the physical activity environmental factors which had an alpha of 0.67. However, this alpha is above the recommended lower level for group comparisons (i.e., > 0.50) [
38]. As the physical activity environmental factors comprised 4 items the mean inter-item correlation is likely a more appropriate statistic for evaluating internal consistency. This measure, like coefficient alpha, produces an index of item homogeneity, but unlike the alpha is not affected by scale length [
39]. For a reliable scale the mean inter item correlation should ideally be within the range of 0.20–0.40. However, values in the range of 0.10 to 0.50 are acceptable [
39,
40].
In addition, table
8 shows comparisons of psychometric properties of scores from the translated measures with those from the original measures [
17].
Table 8
Comparing Current study to Original study for reliability estimates of physical activity related psychosocial scales (n = 512)
Variable | items | alpha | alpha |
Change strategies | 15 | 0.78 | 0.88 |
Self-efficacy | 06 | 0.84 | 0.76 |
Pros | 05 | 0.81 | 0.81 |
Cons | 05 | 0.69 | 0.53 |
Family Support | 04 | 0.72 | 0.79 |
Friend Support | 05 | 0.77 | 0.60 |
Environment | 04 | 0.67 | 0.42 |
Test-retest reliability
Almost 20% (93 subjects) of the original sample (512) were randomly selected to complete the physical activity-related psychosocial measures again 15 days after the initial assessment. Pearson Product Moment Correlations were calculated between the Time 1 and Time 2 assessments for the five scales. Results showed that the relationships were in the large effect size range for scales of physical activity self-efficacy, physical activity social support, physical activity pros and cons, physical activity change strategies and physical activity environmental factors, respectively (0.68, 0.55, 0.40, 0.74, and 0.38).