Writing about Statistical Methods
Example 1: Table 1
Descriptive data are shown as median (range) or n (%), as appropriate. What numbers are shown in the cells of the table?
Continuous variables were compared using the Kruskal-Wallis test. Categorical variables were compared using the \chi^2 or Fisher’s exact tests, as appropriate. How were p-values generated?
Example 1: Table 2+
- Hospital mortality and use of extracorporeal membrane oxygenation were modeled using logistic regression. What outcomes were modeled and how?
Example 1: Table 2+
- Regression results are shown as odds ratio (95% confidence interval; CI). What are in the cells of the table?
Example 1: Table 2+
- Significance of predictors was assessed using the omnibus Wald \chi^2 test. When appropriate, pairwise comparisons were made using the Bonferroni adjustment. How were p-values generated/evaluated?
Example 1: Table 2+
Note the informative footnote!
- I try to include information in footnotes as a reminder for the reader.
Example 2: Table 2+
- Sometimes we decide to show results for a subset of predictors.
Example 2: Table 2+
We just need to make sure that information is noted somewhere.
- Although the information is in the footnote, the models being compared/contrasted must be clearly outlined in the methods section.
Example 3: Table 2+
- Correlations between covariates at initial survey were assessed using Spearman’s correlation. What method was used to assess correlations?
Example 3: Table 2+
- Let’s take a look at the table description and footnotes…
Example 3: Table 2+
- Hierarchical linear models (HLM) were used when modeling all outcomes of interest. First, the models had only sex, type of SCD, and the baseline sleep quality score as predictors. Then, the models were further ajdusted for depression score. Finally, the models were further adjusted for pain frequency, intensity, and interference. Because all HLM incrementally increased the R2 value, only the fully adjusted models are shown.
Example 3: Table 2+
- Sleep quality and age appropriate sleep were modeled using binary logistic regression while sleep duration was modeled using ordinary least squares regression. Regression results are shown as odds ratio (95% confidence interval; CI) for logistic regression and unstandardized beta coefficients (95% CI) for linear regression. Significance of predictors was assessed using the omnibus Wald \chi^2 test for categorical predictors. As necessary, pairwise comparisons were made using the Bonferroni adjustment.
Example 5: Writing More About Methods
- Depending on the field, more information about methods may be necessary to help the reader interpret the results. Consider this graph:
Example 5: Writing More About Methods
- What did the resulting statistical methods section look like?
