3.1 Measurement of physical activity
Measuring physical activity through surveys is complex. To understand the extent to which national guidelines are met, information is required on three key dimensions of activity:
- Intensity: effort required to carry out activity (low, moderate, vigorous)
- Duration: length of time activity carried out (usually in minutes)
- Frequency: number of sessions over a fixed period (per week)
In the SHeS, the intensity level of activities mentioned by participants was estimated to help assess adherence to the physical activity guidelines. The four categories of physical activity 'intensity' were vigorous/moderate/light and inactive. As the guidelines refer to moderate or vigorous activity, only activity reported in these categories were included in analysis of meeting physical activity guidelines. Further details on how intensity of different activities within each of the domains of physical activity assessed in the Scottish Health Survey can be found in the Scottish Health Survey 2012 - Volume 1 Main Report and supplementary technical report.
Physical activity can also take place in a number of different contexts or domains. The Scottish Health Survey asks respondents about their physical activity in five main domains which make up the physical activity measurement:
- Activity at work (occupational physical activity)
- Manual/gardening/DIY work
- Sports and exercise
For the last four domains, survey respondents are asked to report any activities that lasted at least 10 minutes and the number of days in the past four weeks in which they had taken part in such activities. For walking, participants are also asked on how many days they had taken more than one walk of at least 10 minutes. Where a participant has taken more than one walk, the total time spent walking for that day was calculated as twice the average reported walk time.
It is worth noting that the walking domain in the SHeS includes walking for recreational purposes and walking for transport and so, while it is the closest domain to active travel, it does not completely map onto this physical activity context.
Further, in response to concerns that the method for grading the intensity of walking was underestimating older adults' exertion levels, an additional question on walking was asked of those aged over 65 in the 2012 survey:
During the past four weeks, was the effort of walking for 10 minutes or more usually enough to make you breathe faster, feel warmer or sweat?
The overall impact of this addition on physical activity estimates for all adults was shown in the 2012 Health Survey Report to be minimal. The addition did, however, affect older age groups in the way expected - a higher proportion of older age groups met the 2012 guideline. It is possible this change could contribute to differences in patterning of physical activity between 2011 and 2012.
To analyse patterning of factors in relation to the different domains of physical activity, derived binary outcome variables were created for each of the domains (any/no participation).
Occupational physical activity was calculated in a slightly different way in this study than the approach in the 2012 Scottish Health Survey. The 2012 Scottish Health Survey used an updated definition of occupational physical activity which combined information on intensity of activity carried out in work with a new question on sedentary behaviour in work to produce estimates of the duration of moderate activity at work per week. In this study, a respondent was classified as participating in occupational physical activity if they reported being either very or fairly physically active in work. Those reporting low or no physical activity in work were classified as non-participants. This simpler methodology was used because our interest was to understand factors associated with any participation in activity in the respective domains.
3.2 Logistic regression
Regression analysis was used to explore whether or not various demographic, socioeconomic and health/lifestyle variables were independently associated with (a) meeting the physical activity guidelines in 2011 and 2012 and (b) any participation in the different domains of physical activity.
Logistic regression is a statistical technique that enables examination of the relationship between a dependent variable (in this case, either meeting the physical activity guidelines or participation in activity in different domains) and various independent (or predictor) variables (sex, age, income, health status etc). The analysis identifies which of these independent variables are significantly and independently associated with the dependent variable after controlling for inter-relationships between the variables. The analysis also gives an indication of the relative strength of different factors.
Logistic regression models the log 'odds' of a binary outcome variable (for example, the odds of meeting the physical activity guidelines compared to not meeting them). The odds ratio is a measure of the likelihood of the outcome for one group compared to another group.
Odds ratios describe the strength of association between two binary variable values. For example, if being young has an odds ratio of 2, it means that the odds of achieving the recommended physical activity levels are two times higher in those who are young compared to those who are older, when all the other variables in the model are held constant.
Bivariate analysis (cross-tabulation) was conducted on 2012 data for meeting the physical activity guidelines with a range of demographic, socioeconomic and health and lifestyle factors identified for investigation based on the literature, see Table 1.
|Health and lifestyle
|Disability (long-standing illness)
|Economic activity status
|Level of Education
|Mental Well-being (MWB)
* More detail on variables used in the bivariate analysis is available in Table 9 in Appendix A.
Although there is evidence indicating a possible impact of ethnicity and religion on physical activity, this present study is of a single year's survey data, and the sample sizes for individual ethnic and religious groups were too small to include in any meaningful analysis.
As noted in Chapter 2, environmental factors are important in relation to physical activity. Urban/Rural classification and Health Board were tested and found to be non-significant and so were excluded from further analysis. Other environmental factors were not tested because relevant variables were not present in the SHeS.
Chi-squared tests of association and significance level were performed and results are presented in Chapter 3. Multivariate logistic regression models were subsequently created to examine the relationship between the range of factors and the likelihood of meeting the physical activity guidelines in 2011 and 2012. Only variables that were significantly associated with physical activity outcomes in the bivariate analysis at the 95% level were included in the logistic regression models. Separate models were run for 2011 and 2012 to examine the effect of the guideline change.
A series of logistic regression models were also run to investigate factors associated with participation in the different domains of overall physical activity. Each of the regressions were repeated separately for men and women, as the literature suggests gender-specific differences in the factors associated with physical activity and meeting the recommended guidelines.
The analysis identifies which of the independent variables (e.g. income or marital status) are significantly and independently associated with the dependent variable (physical activity outcome), after controlling for inter-relationships between the other variables in the model. Collinearity (the association between two or more predictor variables) was tested and no relationships of sufficient strength were identified to exclude any of the predictor variables from being entered into the model. Starting with a basic model containing just sex and age, each variable was added into the model using stepwise regression analysis in SAS. The models were run without constraints on the level to include or exclude each individual variable and changes in the odds ratios with each additional variable included were monitored as further check on potential collinearity. This stepwise method combines forward and backwards selection and allows identification of variables by order of association strength (based on chi-squared score). By using stepwise analysis it is possible to explore how adding new variables reduces the effect of previous variables, and also shows the increased/decreased explanation value (R2), which explains how much the outcome is based on each variable, and overall for a combination of the predictive variables. Total model R2 values are presented in Table 12 in Annex B showing the amount of physical activity outcome variation explained by each model.
The chi-squared scores for each of the predictor factors entered into the model were used to indicate which characteristics tested in this analysis had the highest relative influence on meeting the physical activity guidelines and to determine the order of variable entry into the model. For example, if the health predictor had a higher chi-square value than sex or age predictors, the health variable was entered first, and the chi-squared value would be recalculated for the remaining two variables. Odds ratios were calculated to show the odds of a category within a variable occurring compared to a specified reference category. All analyses were tested at the 5% significant level producing 95% confidence intervals.
All analysis is based on complete cases. No multiple imputation was used. Of the eligible sample of adults aged 16 and over, only two respondents were excluded through missing a physical activity outcome value and 25% were excluded from the main logistic regression due to missing predictor values.
3.3 Limitations of the analysis
The Scottish Health Survey relies on self-report of physical activity, which has well-recognised limitations in accurately assessing physical activity due to difficulty in accurately recalling physical activity, differences in perceptions of physical activity intensity and matching responses to what is perceived to be the societal norm (social desirability responding). The advantages of using self-report are that it is easy to collect data from a large number of people at low cost, many survey instruments have been validated against more accurate methods and repeatedly used in research. Self-reported physical activity is widely used and, despite the issues with absolute accuracy, this research allows comparison to a large body of prior evidence. In addition, the SHeS calculates total physical activity based on questions about different domains of physical activity. This more specific approach to assessing total physical activity may be expected to reflect more accurately total physical activity.
The study is limited to analysis of variables that are in the SHeS. A wide range of other factors, including environmental, psychological and interpersonal ones have been shown to have a relationship to physical activity outcomes, as already outlined in Chapter 2. Thus any model of explanation produced from the analysis in this study will only be partial.
Finally, many of the factors examined in the models for this study are likely to have bidirectional relationships with physical activity. For example, if a relationship was found to exist between having a BMI of >30+ (obese) and meeting physical activity guidelines, it could be said that not meeting physical activity guidelines is associated with being obese, but equally that being obese is associated with not meeting the guidelines. No clear direction of causality can be claimed.
Email: Niamh O'Connor
There is a problem
Thanks for your feedback