Data Analysis
A power analysis was performed for an analysis of variance with two levels, a medium effect size d = 0.75 and α = 0.05. For a power (1 – β) of 0.8, a minimum of 30 subjects per group were required for the planned analysis. At the conclusion of the four week period, 1,409 subjects had accessed the survey. Because the dependent variable, SOP, was critical to all of the planned analyses, data sets that were missing this value were removed, leaving 1,202 usable data sets available for analysis, which far exceeded the sample size indicated by the power analysis. A missing data analysis revealed no pattern to missing data.
All data analyses were conducted with SPSS version 19.0 (SPSS Inc, Chicago, IL). Sample demographics were compared with the population using chi-square analysis for categoric variables and the t test for the continuous variable (Table 1).
The dependent variable was tested for normality and homogeneity of variance. The SOP-VAS scores displayed a non-normal distribution with a left skew. However, when using samples greater than 200, a variable with statistically significant skewness often does not deviate enough from normality to make a substantive difference in the analysis. Additionally, 1-way ANOVA is robust to violations of normality, with only a small effect on the type I error rate. Because the nonparametric Kruskall-Wallis test can also be used on non-normal data to compare independent groups, both the Kruskall-Wallis test and ANOVA were performed on the data to confirm statistical significance. Because ANOVA tolerates violations of normality of the dependent variable, it is reported because it is the preferred test due to the interval level of data being analyzed. When large disparities were found in the number of subjects in each group, the Brown-Forsythe test was performed. This test is also preferred with heavily skewed distributions. The Results section clearly indicates the test reported for each category.