Sample size

One of the most important issues to be considered early in the process of planning a study is the size of the sample required. The sample size decided upon will determine (a) the precision of the resultant estimates and (b) the power of a significance test to detect a difference between groups or an association between variables should one actually exist. In determining the sample size, a researcher must balance two competing demands: the need for the study to be sufficiently large to have a reasonable chance of answering the research questions, and the need for the study to have minimum cost, in terms of actual dollar cost and experiment costs (e.g., participants’ time and inconvenience).

The proposed sampling design for studies looking at the association between environmental attributes and physical activity behavior is multistage. Firstly, neighborhoods that match specific physical and social characteristics are selected. Secondly, residents are randomly selected from these neighborhoods. This type of sampling scheme includes two kinds of sample sizes: the sample size of the macro-units (N; neighborhoods) and the sample size of micro-units (n; residents) within each macro-unit, with N x n being the total sample size for the micro-units. We need to establish the two sample sizes that will yield an acceptable level of statistical power (0.80) to detect individual-level (e.g., what is the relationship between perceived access to shops and walking for transport?) and individual-neighborhood cross-level relationships (e.g., what is the relationship between the average neighborhood street connectivity and walking for transport?) with minimal cost.

The suggested samples sizes were empirically derived using data from the NQLS and PLACE studies. All analyses were performed using PINT, a program for power analysis in studies with a two-level design (Bosker, Snijders, & Guldemond, 1999). The table below reports the sample sizes needed to achieve a power of at least 0.80, assuming that the individual-level and neighborhood-level predictors each explain 1% of the variation in the outcome and that the intraclass correlation for the outcome is 0.05. The table shows that, because the expenses of a study are in the main determined by the total sample size, it is more cost-effective to aim for a large sample of neighborhoods (if possible) than a large sample of residents within neighborhoods.

* For the purpose of this power analysis, neighborhood was defined as a census collection district.

Age range:

Priority 1: 18-65 years. The current studies and the IPAQ are designed for this age group.

Priority 2: 65+ years. The environmental variables should be the same or similar as for younger adults. The CHAMPS is the proposed physical activity measure for older adults.

Priority 3: 4-18 years. Preliminary work is needed to identify candidate environmental variables and develop appropriate measures. Objective physical activity measures, such as accelerometers, will need to be used with this age group. Investigators interested in youth will need to form a planning group for a future study.

Sample size per country: approx. 500 per age group (i.e., younger adults, older adults), balanced by gender.

If specific neighborhoods are studied: approx 50 per neighborhood.

Other Participants Information