Table 1 Comparison of traditional and mixed-effect approaches and questions to answer when choosing a statistical method for the analysis of labour pain in repeated measures data
From: Labour Q1 pain – poorly analysed and reported: a systematic review
Questions to ask when you choose statistical method | Statistical property | Statistical method | ||
|---|---|---|---|---|
End-point Analysis | rANOVA | Mixed effect models | ||
1. What do you want to compare? | Research question | Compares mean labour pain between groups at one time-point | Compares mean labour pain between groups at several time-points Study interactions between timeatreatments | a. Compares mean labour pain between groups at several time-points b. Study interactions between time and treatment c. Study individual women’s pain changes over time |
2. Do you have measurements of labour pain at all time-points for all women? | Missing data | Excludes woman with missing measurements | Excludes woman with missing measurements | Use all available measurements under the assumption of missing at random (MAR) |
Possible effect of omitting women with missing values? | Sample bias | Sample bias | Not applicable | |
Possible effect of imputation of missing data? | Estimation bias | Estimation bias | Not applicable | |
3. Can you assume that correlation of pain is equal between all time-points? | Assumption on the between woman pain correlation | Independent | Independent | Independent |
Are the labour pain assessments made with unequal distances, e.g. at baseline, and after 2 and 6 h? | Assumption on the individual woman pain correlation or covariance matrix | Independent | Compound symmetry | Allow a variety of covariance structures, e.g. Independence, Compound symmetry, AR [1], Unstructured |
4. Can you assume that the variance of labour pain is equal at all time-points? | Assumption on the variance of pain at different time-points | Constrained to be equal at all time-points | Constrained to be equal at all time-points | Allowed to vary |
5. Are measurements of labour pain normally distributed? | Assumption of normal distribution | Normality assumption | Normality assumption | Normality assumption |
6. What requirements do you have on your statistical model to model the pain over time? | Description of time effect | Simple | Flexible | Flexible |
7. Would you like to consider labour pain traits for individual women over time? | Estimation of individual trends | No | No | Yes |
8. Do you need to adjust labour pain for factors that vary during labour, e.g. cervical dilatation and use of other pain relief? | Time dependent covariates | No | Yes | Yes |
9. Do you have knowledge of applied statistics? | Ease of implementation | Very easy | Easy | Complex |
10. Do you have access to a good computer? | Computational complexity | Low | Low | High |