docs/api: revise methods for gee glm, part of #1 & #2

docs/api/methods.tex

@@ -161,9 +161,12 @@ \subsubsection{ART Prevention Impact}
 relative reduction in incidence and/or proportion of infections averted
 relative to the base-case scenario for available time horizons relative to \tx{0}.
 \par
-Finally, for each factor of heterogeneity,
+For each factor of heterogeneity,
 we compared projected ART impacts (incidence reduction/infections averted)
 across different factor levels (whether or not, and how the factor was modelled).
-We plotted the impacts versus time since \tx{0}, stratified by factor levels,
-and explored whether projected impacts were the same under all factor levels,
-testing for significant differences using a Kruskal-Wallis test.
+We estimated the effect of each factor level on ART impacts
+using linear multivariate regression, with generalized estimating equations \cite{Hojsgaard2006}
+to control for clustering due to multiple estimates per study.
+Time since \tx{0} was included as a covariate, and two variables were removed due to missingness.
+No variable selection was used to avoid biasing effect estimates \cite{Harrell2001}.
+We also plotted impacts versus time since \tx{0}, stratified by factor levels.