A backward stepwise method was used whereby, at each step, variables with a p-value of >0.1 on the likelihood ratio test were removed.
In order to minimise the amount of missing data for each item, a 'best estimation' method was used whereby respondents with missing data were assigned the item-score of a respondent with a similar age, gender, income, and education level [ 37- 39].
A rolling selection method was used, whereby each drug claim was evaluated on the criteria described below and the earliest claim meeting all criteria was selected as the index date and the medication class (DPP-4i, SU, or TZD) was considered the index drug class.
To maximize the accuracy of the analysis (in particular, to ensure documentation of the number of subjects monitored during the week) the zero-reporting method was used, whereby doctors were also required to report the absence of cases among the patients.
For the training sample, prediction of class membership was based on the internal voting scheme for RF, whereas for PLS-DA a softmax method was used, whereby the predicted class, c∗, is the largest class probability after model predictions have been transformed on a [0,1] interval (with unit sum), that is (5) Y = y c where c ∗ = arg max 0 ≤ Z c ≤ 1, ∑ Z c = 1 (Z ^ c ).
Consistent with grounded theory methods, theoretical sampling was used whereby data collection was directed by concepts arising in the evolving theory [ 14, 16].
14 This might be due to the method of questionnaire administration used, whereby each respondent was given a pen and a sealed envelope in which to return the survey, in order to preserve confidentiality and anonymity; this suggests that the technique was effective in increasing response rates.
The comparative Ct method for relative quantification was used, whereby all threshold cycles (Ct) are first normalized to the expression of an endogenous control (18s rRNA).
An incomplete fixed block experimental design was used, whereby each analyst used one method for each job.
What are the methods used?
