Objectives: To develop a valid and reliable internet based lifestyle physical activity questionnaire suitable for use among the United Kingdom population.
Methods: After a detailed content analysis and item generation using a panel of experts, an internet based measure of lifestyle physical activity behaviour was developed. Data were collected from 1369 subjects in total. Confirmatory factor analysis was used to examine the two subscales of the Brunel lifestyle physical activity questionnaire among independent samples and by use of multisample analyses.
Results: The confirmatory factor analysis showed the psychometric integrity of two subscales: planned physical activity and unplanned physical activity.
Conclusion: The questionnaire is a valid and reliable instrument designed to provide an online behavioural assessment to be used in conjunction with a 12 week personalised fitness programme delivered through the internet.
Statistics from Altmetric.com
- BLPAQ, Brunel lifestyle physical activity questionnaire
- CFA, confirmatory factor analysis
- CFI, comparative fit index
- EFA, exploratory factor analysis
- PA, physical activity
- PPA, planned physical activity
- SRMR, standardised root mean residual
- UPA, unplanned physical activity
Measurement of lifestyle physical activity (PA) by self administered behavioural assessment facilitates effective screening, monitoring, and intervention.1 It is necessary for PA interventions to move towards mass media approaches that make more effective use of newer technologies such as the internet.2 The Brunel lifestyle physical activity questionnaire (BLPAQ), the questionnaire developed in this study, provides one way in which to exploit the power of the internet to improve people’s health.
The theoretical underpinnings of the BLPAQ reflect the necessity to distinguish between planned and unplanned PA.1 It has been proposed that, to maintain optimal health, people should engage in at least 30 minutes of daily PA that comprises both planned and unplanned modes.3 A number of studies have designed questionnaires to tap PA behaviours without distinguishing between planned and unplanned modes.4–6 Further, concerns with measurement have pervaded previous attempts to assess PA behaviour, typically aspects of the validity and reliability of instruments.7,8 The purpose of the present study was to develop a valid and reliable lifestyle PA questionnaire, designed for use on the internet, suitable for the United Kingdom population.
Our research strategy was to develop a questionnaire and examine its validity in 10 stages. On the basis of conceptual discussions of lifestyle PA,1,9 items were developed to tap the two PA domains of planned PA (PPA) and unplanned PA (UPA). A latent variables analysis approach was used to facilitate inference of overall levels of PPA and UPA from a number of observable “manifest” variables or indicators.10
Stage 1: generation of item pool (expert panel 1)
The generation of the initial item pool was based on a number of considerations including (a) existing questionnaires,4,11,12 (b) our understanding of the nature of lifestyle PA, and (c) input from a panel of experts (expert panel 1). The panel of experts assisted in the generation of items and established the extent to which the initial item pool tapped the intended constructs. Expert panel 1 comprised 12 people (six men and six women; mean (SD) age 36 (5) years) who worked in the health and fitness industry, were academics with an interest in health and fitness, and/or possessed knowledge of questionnaire development. Demographic details are an important consideration for diagnostic purposes.13 Thus a series of demographic items were developed (available on request), which were also scrutinised by expert panel 1.
Stage 2: item comprehensibility and applicability (pilot sample 1)
To establish the comprehensibility and applicability of the items, and to make fine adjustments, the initial item pool of 10 items was piloted among a panel of 16 members of the lay public (six men, seven women, three did not report their sex; mean (SD) age 35 (15) years). This panel comprised a purposive stratified sample intended to reflect a range of socioeconomic groups, different age groups, both sexes, and diverse ethnic minorities.
Stage 3: further test of content validity of the BLPAQ (expert panel 2)
To establish the importance of each item to the measurement of the intended construct, a panel of 36 experts (mean (SD) age 34 (8) years) rated the refined item pool and reworded/deleted/added items as necessary. There were 22 men (mean (SD) age 34 (7) years) and 14 women (mean (SD) age 32 (8) years) in expert panel 2. This group extended the work completed by expert panel 1 and pilot sample 1. Fifteen members of expert panel 2 had a doctoral qualification in a related area and five were full professors. The remainder were educated to at least master’s level and drawn form both industry and academia. Figure 1 shows the final version of the BLPAQ derived from this procedure.
Structure of the BLPAQ
Respondents were asked to provide honest answers about their activity behaviour. The section on lifestyle PA behaviours was preceded by the definition: “planned PA is any activity that is scheduled into your daily routine, which may enhance your health, fitness, or wellbeing.”1 Examples include brisk walking, gardening, cycling, team games, etc. Responses to each item in the PA subscale were provided on a five point continuous closed numerical scale consisting of the following anchors relating to a “normal” week: never, 1–2 times, 3–4 times, 5–6 times, 7 or more times. Participants responded by either ticking or clicking with their mouse. The six items measuring PPA were intended to tap the intensity, frequency, and duration of such activity. Frequency of unplanned behaviour was not assessed given the difficulty in obtaining valid measurements because of its highly transitory nature.14
Stage 4: administration of the BLPAQ in pen and paper format (pilot sample 2a and 2b)
In stage 4, a pen and paper based version of the BLPAQ was administered to 563 volunteers representing a broad cross section of the United Kingdom population in terms of socioeconomic and ethnic background (checked against 2001 UK Census results). The mean (SD) age was 32 (13) years, with 260 men (mean (SD) age 32 (13) years) and 303 women (mean (SD) age 31 (12) years).
Using item scores from the pilot sample 2 data, a univariate outlier test (z > ± 3.29)10 revealed 12 outliers, which were deleted. Twenty one multivariate outliers were also identified and deleted using Mahalanobis’ distance test (p<0.001).10 The remaining 530 cases were split randomly into two equal groups—pilot samples 2a and 2b—with the first half used to explore the factor structure of the BLPAQ.
Stage 5: confirmation of the BLPAQ factor structure
Pilot sample 2b was used to confirm the factor structure of the BLPAQ using results of an exploratory factor analysis (EFA), and to test competing models using structural equation modelling techniques (EQS version 5.7).
Stage 6: piloting the BLPAQ on the internet (pilot sample 3)
An internet based version of the BLPAQ was used to collect data from pilot sample 3: 742 volunteers representing a broad cross section of the United Kingdom population in terms of socioeconomic and ethnic background. The mean age of respondents in pilot sample 3 was 37 (10) years, with 450 men (mean (SD) age 36 (9) years) and 292 women (mean (SD) age 37 (10) years). After initial analyses, eight univariate outliers and 15 multivariate outliers were deleted. Pilot sample 3 was used to reconfirm the factor structure of the BLPAQ and to test its invariance with the data derived from pen and paper administration in stage 5. An additional item was introduced to identify types of physical activities that respondents most enjoyed, which could be fed back within the internet based intervention programme.
Statistical methods for stages 4–10
EFA and confirmatory factor analysis (CFA) were used to examine the validity of the factor structure of the BLPAQ. In stage 4, EFA was used on data from pilot sample 2a (n = 265) to test for the hypothesised PPA and UPA factors. In stage 5, pilot sample 2b (n = 265) was used to confirm the hypothesised factors and to test the tenability of competing models that were theoretically meaningful. In stage 6, data from pilot sample 3 (n = 719) were used to confirm the factor structure using internet based completion. Owing to the change in data collection medium, competing models were re-examined.
The model was estimated using the maximum likelihood estimation method, as Mardia’s normalised estimate (pilot sample 2a = 2.09) indicated that the data were not multivariate normally distributed. The comparative fit index (CFI) and standardised root mean residual (SRMR) were used to assess model fit. These are purported to out-perform other goodness of fit statistics.15
According to Hu and Bentler,16 the cut-off value relating to a relatively good fit between the hypothesised model and the observed should be close to 0.95 for the CFI, and close to 0.08 for the SRMR. These indices were used to evaluate model fit. In addition, Akaike’s information criterion was used to test the tenability of competing models. The model with the lowest Akaike’s information criterion is considered to have the best fit.
In stage 7, the invariance of the factor structure was tested using pilot sample 2b and pilot sample 3 with multisample CFA. Before this procedure, the fit of the model was tested independently with the hypothesised two factor model. In stage 8, the standardised solutions for each sample were examined.
In stage 9, having tested for the relevant parametric assumptions,10 a three way independent samples multivariate analysis of variance was used to examine subgroup differences with the Statistical Package for Social Sciences (SPSS) version 10.0: 2 (groups: pilot sample 2 and pilot sample 3) × 2 (sex) × 4 (age groups). The data were split into four equal age groups: 18–27 years (n = 253); 28–34 years (n = 266); 35–42 years (n = 232); 43–73 years (n = 233). Recommendations for BLPAQ norms were made on the basis of the multivariate analysis of variance results.
In stage 10, the internal consistency of the two factors was assessed using Cronbach α coefficients.
The results are presented sequentially in accordance with the methodological stages described above. Given that stages 1–3 involved the generation of items and the establishment of content validity, the presentation begins with stage 4.
Stage 4: EFA
Table 1 contains the results of the EFA. Principal components analysis extracted two factors accounting for 64.6% of the variance. A clear factor solution emerged with strong loadings (>0.60) on each of the two expected factors of PPA and UPA. This factor structure was consistent with theoretical predictions.1
Stage 5: CFA
Based on the EFA results (table 1), two factors were hypothesised to emerge from the lifestyle PA items: PPA and UPA. Consequently, a nine item, two factor model was tested using CFA on the data obtained from pilot sample 2b. Overall, the data from pilot sample 2b showed an acceptable fit (table 2), with the robust CFI (0.94) very close to the criterion value.16
Stage 6: test and re-test of competing models
The validity of a competing one factor congeneric model, based on the hypothesis that participants did not distinguish between PPA and UPA, was tested using CFA (table 2). The goodness of fit indices for the congeneric model showed a poor fit to the data (all indices <0.90). The results showed better fit indices for the two factor model, providing evidence of discriminant validity for PPA and UPA. In addition, we tested a Tau equivalent model, which has equal true score variances and equal error variances, and results indicated a poor fit. Alternative models were re-tested using pilot sample 3 (table 3).
Stage 7: multisample CFA
Given that there was a strong fit in pilot sample 2b and pilot sample 3 independently for the two factor model, it was hypothesised that factor loadings would be equal across these samples. Thus the Lagrange multiplier test was used to assess whether equality constraints were imposed correctly. Multisample CFA, with factor loadings constrained to be equal across both samples, indicated a CFI of 0.91 (table 4), which is slightly below the criterion value.16
The Lagrange multiplier test results indicated that three items had significantly different factor loadings across samples. Specifically, releasing the equality constraint for: (a) the intensity of the UPA item would reduce χ2 by 51.1 (v = 5, p<0.01); (b) the duration per session of the PPA item would reduce χ2 by 42.6 (v = 4, p<0.01); (c) the intensity of the PPA item would reduce χ2 by 23.9 (v = 2, p<0.001). Subsequently, the CFI increased to 0.92 (SRMR = 0.07).
The Lagrange multiplier test supported the notion that constituents of PPA and UPA should be allowed to correlate. The correlation between the two factors was significant (r = 0.10, p<0.01); however, given that significant findings are boosted by large sample sizes10 and the percentage of variance explained is only 1.1%, the factors are largely orthogonal in nature.
Stage 8: standardised solutions
When acceptable fit indices have been evidenced, it is appropriate to examine the standardised solutions of a sample to assess the amount of unique variance accounted for each item by the factor (table 5). Table 5 indicates that all of the items tap unique variance other than the item concerning total time engaged in PPA, which exceeded the cut-off point for error variance of 0.90.
Stage 9: subgroup differences
Subgroup differences were examined using combined data from pilot sample 2b and pilot sample 3 to determine whether the BLPAQ would require separate sets of norms. A three way multivariate analysis of variance of BLPAQ factor scores by sample, age, and sex (table 6) indicated a significant multivariate interaction effect for sample by age group (Wilks’ λ = 0.99; F6,1934 = 2.13; p<0.05; η2 = 0.007). Follow up univariate analyses showed that the interaction effect held only for PPA (F3,968 = 3.05; p<0.05; η2 = 0.009); however, Tukey’s post hoc test with Bonferroni adjustment did not indicate any significant differences. The multivariate analysis of variance also revealed main effects for sample (Hotelling’s T = 0.009; F2,967 = 19.60; p<0.05; η2 = 0.039) and sex (Hotelling’s T = 0.009; F2,967 = 4.46; p<0.05; η2 = 0.009). Follow up univariate analyses for sample (PPA: F1,968 = 19.77, p<0.001, η2 = 0.020; UPA: F1,968 = 23.14, p<0.001, η2 = 0.023) revealed that the pen and paper sample (pilot sample 2b) reported significantly higher levels of both PPA and UPA when compared with the internet sample (pilot sample 3). Across both samples, women reported that they engaged in more UPA than men.
In all cases, the significant differences were associated with very small effect sizes as indicated by the η2 statistic.18 Therefore, separate tables of norms for each subgroup of the population are not required.
Stage 10: internal consistency
Internal consistency estimates for the BLPAQ subscales using Cronbach’s α were as follows for PPA and UPA respectively: pilot sample 2b, α = 0.91 and 0.65; pilot sample 3, α = 0.88 and 0.68; and both samples combined, α = 0.90 and 0.68. UPA has a marginal α coefficient, which did not exceed the cut-off criterion of 0.70.10
Two subscales of the BLPAQ were identified using EFA and supported by a series of CFAs. Tests of alternative models revealed that the two factor solution comprising PPA and UPA was the most stable. The multisample analysis (pilot sample 2b v pilot sample 3) showed that three items required equality constraints to be released, which resulted in a marginal fit (CFI = 0.92, SRMR = 0.07). Detailed analysis indicated that there was slight instability between samples in how respondents perceived the intensity of their UPA, the duration of their PPA per session, and the intensity of their PPA. This instability may reflect difficulties in summating the information requested by the items as it may vary from day to day. It is not possible to identify the precise source of this variation; however, secondary analyses showed that, for all three items, significantly (p<0.01) higher scores were reported by pilot sample 2b. One plausible explanation is that, if this sample were on the whole more physically active, they would be able to recall their PA habits with greater accuracy.17
What is already known on this topic
For optimal health, people should engage in at least 30 minutes of daily physical activity comprising both planned and unplanned modes. A number of questionnaires have been designed to draw on PA behaviours without distinguishing between planned and unplanned modes.
In addition to slight instability in the factor loadings between pen and paper and internet based versions of the BLPAQ, there were clear differences in the nature of the data collected. Most notably, it appeared that respondents reported engaging in significantly more PPA and UPA in the former. One plausible interpretation is the occurrence of social desirability given that respondents were actively recruited. In addition, internet based respondents were seeking to increase their PA and wanted to use the questionnaire as a vehicle towards this end.
Regardless of the mode of data collection, women reported more UPA than men (table 6). This may indicate that, in the United Kingdom, women are more likely to engage in activities such as housework, walking the dog, shopping, and playing with children.18 Further, it is interesting to note that the UPA reported by women is significantly greater (t982 = −4.55; 97.5% confidence interval = −0.34 to −0.12; p<0.001) than the proportion of UPA engaged in by men (female mean UPA = 43.2%; male mean UPA = 39.8%).
Strengths and limitations
The use of health behaviour change programmes employing internet based technologies is cost effective and allows participants to work at their own pace and convenience.19 Home based programmes enhance accessibility for people limited by finances or transportation.20
Potential disadvantages of internet based methods of communication are that a high financial outlay is required to develop systems such as the one described herein. Also, access to internet based programmes is not yet universal. In December 2003, 49% of United Kingdom households had internet access (http://www.statistics.gov.uk/releases). Further, instruments such as the BLPAQ should not be administered as a substitute for professional medical support, but as a complement to it.
A limitation with the structure of the BLPAQ concerns the marginal α of the three item UPA subscale. In cases where the number of items in a subscale is less than 10, an α coefficient of 0.60 is acceptable as long as there is evidence for validity and there are good theoretical and/or practical reasons for the subscale.21 Further, the homogeneity of this subscale was demonstrated by CFA, which is a more rigorous test than Cronbach’s α, and there is sound theoretical premise for the content of the PPA subscale.
What this study adds
A valid and reliable internet based questionnaire has been designed which allows researchers to test theories underlying planned and unplanned physical activity behaviour.
Conclusions and recommendations
The BLPAQ is a valid and reliable internet based questionnaire that allows researchers to test theories underlying PPA and UPA behaviour.22 A major recommendation is for extension of the validation process to test concurrent and predictive validity. Further, if unplanned daily activities at a moderate intensity can be promoted, it is more likely that PA requirements can be met. The notion of integrating unplanned lifestyle PA behaviours to enhance health status is concordant with current thinking among exercise professionals, government agencies, and epidemiologists.1,9,23
We thank Biomedical Research Ltd and the Brunel Research Committee for supporting this research. The online version of the Brunel lifestyle physical activity questionnaire (BLPAQ) can be viewed at the following location with a username and password available from the first author: http://www.slendertonementor.com.
Present address: N L D Chatzisarantis is now at the School of Sport and Health Sciences, University of Exeter, Exeter, Devon, UK
Competing interests: The research was funded in part by Biomedical Research Ltd, Republic of Ireland. The questionnaire reported in the study is used as part of an on line health and fitness programme. Although the research was sponsored by a commercial organisation, this had absolutely no bearing on the results.
Ethics: The study received ethical approval from the institutional review board and written informed consent was sought from subjects.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.