Questionnaires

In many research questionnaires are being used to get information from respondents. This can be by paper, but also electronically. Often concepts and measurement scales are being used, but it can also be a simple questionnaire in which it is asked how a student have chosen for a certian school, what his background is etc.

In all those cases it is good to analyse these questionnaires. This can be done with simple straight counts, followed by cross tables. If concepts have been measured with a set of items, it is desirable to check whether the concepts to be measured have a certain consistency and / or these concepts measure what they should measure (validate). Various methods can be used to determine whether there is internal consistency (e.g. with Cronbach's alpha) or that certain dimensions can be found in the concepts (e.g. with factor analysis or HOMALS-like methods).

When all concepts have been measured, the next step is often to make connections. Often times it is most convenient to start with simple cross tables, possibly statistically tested for coherence (eg Fisher Exact test, McNemar test, Chi-square test). Sometimes it is also possible to go a step further by creating a predictive model that tries to measure a response variable by a set of measured factors / explanatory variables. Depending on the design of the study, the measurement level of variables and the amount of observations, can then be chosen for a specific model (e.g. linear regression analysis, ANOVA, ANCOVA, logistic regression analysis, Cox Proportional hazards regression, mixed effect models, repeated meausurements ANOVA, etc.).

Is there rather a situation in which there is a response variable in which it must be checked whether there are differences between two or more groups (e.g. two drugs), then the standard tests can often be used for randomization (experimental design). However, if it concerns an observational study, then possible confounding will have to be taken into account, for which a propensity score analysis will be a good choice. On this subject dr. E.P. Martens wrote his thesis in 2007.

privacy

disclaimer