Using bayesian statistics to interpret non-significant findings from an intervention aiming to promote physical activity to people with spinal cord injury


Background: Evaluations of physical activity interventions for persons with disabilities are often limited by small sample sizes, thereby lacking statistical power. In classical null-hypothesis testing, the absence of evidence of a significant effect (i.e. p>.05) is not evidence for the absence of an effect (i.e. support for the null hypothesis). Bayesian statistics can aid in determining whether non-significant findings support the null hypothesis or indicate data are insensitive. Purpose: Demonstrate the utility of Bayesian statistics for interpreting non-significant findings using data from an intervention that aimed to promote physical activity to people with spinal cord injury. Method: 25 participants with a spinal cord injury (Mean age= 41.36±11.52) were randomized or allocated to an experimental or control condition. Physical activity cognitions were measured pre and post intervention. Apart from social support, significant differences were not detected between conditions (ps>.05). Bayes factors were used to interpret non-significant findings. A Bayes factor is a ratio of likelihood and represents the strength of support for the alternative hypothesis (H1) relative to the null (Ho). A Bayes factor >3 indicates support for H1, o, and .33-3 insensitive data. Results: For all non-significant findings, Bayes factors were between .33-3. Discussion: Bayes factors did not indicate support for the absence of an intervention effect (i.e. the data are insensitive). Bayesian statistics helped to address limitations of classical statistics for interpreting non-significant results and indicated that further research examining the effectiveness of the intervention is warranted.