In a same research thesis can we use hypotheses which are relevant to the study but require different statistical tests for analysis? for example in a study about media consumption and its effect on pro-social behavior, if one hypothesis is about relationship between time spent on watching TV (continuous variable) and Pro-social behavior (measured in time spent with people) (continuous variable). if we use correlation or regression for this and in the same study if another hypothesis is about comparison among exposure to different kinds of media like TV, radio, social media etc (nominal variable measure) and its effect on pro-social behavior (continuous) and we use t-test or ANOVA for testing this hypothesis, is it right? and what if we include both parametric and non-parametric tests in a same study according to the requirement of hypotheses and data. can we do that? or we have to make our hypotheses in a way so that all of them can be tested by same or related statistical tests? is there any rule regarding it? sometime in a large study when we break down the independent and dependent variables then some sub categories could be continuous and some may be categorical. The whole point here is that can we combine different kinds of tests (according to the need of data and hypotheses) in one study.Thanks.
If you study a variable (like "pro-social behaviour") with respect to its dependency on several different possible influencing factors (like "time spend watching TV" and "time spend with people", and "age", and "gender", and "education", and...) I would suggest to model the response variable ("pro-scocial behaviour") as some function of all the other factors, possibly including interactions (since it might be, for instance, that "time spend with people" has a large impact on "pro social behaviour" for females but not so for males, which is represented by and interaction between "gender" and "time spend with people". Such relations can be studies in a sensible way only if the response variable is at least interval sclaed. Using ordinal response variables (as often obtained as psychometric measures, e.g. Likert data) may give very misleading results. Therefore, if this would be the reaserch goal one must take care to operationalize the response variable as at least interval scaled. If this was not done, then the whole analysis really is a "post-mortem analysis" (you can not really do any good analysis because the data is not suited to perform any good analysis; whatever you can do in this case is not ideal)
A diffent issue is it when you analyze different experimental impacts on your response variable in different, isolated experiments. For instance, in one experiment you test the impact of "listening 2h to radio" against "watching 2h TV" on some subsequent reaction of the subjects; in another experiment th impact of,say, light (intensity, color, or quality) on the subsequent "social behaviour" of the subjects was tested; in a third experiment it was tested how different surroundings (home, office, public place, pub, ...) influences the subsequent "social behaviour" of the subjects etc. There is no connection between these experiments, and it is straight-forward to analyze them seperately, with any appropriate method. Since the response variable is the same in all cases, all chosen methods should be suited for this variable (its scale, like nominal, ordinal, interval and -for interval-sclaed variables- the assumptions about the distribution should be sensible). It depends on the kind of the response variable and on the particular kind of hypotheses under consideration if parametric or non-parametric models will be in place. It does not make much sense to mix these approches in different analyses of the same response variable.
I would not focus on testing but on estimation.
The quick answer is yes you can use different tests to answer the same hypothesis. However, this answer contradict basic principles in research, that is the consistency in the different parts of the study should be maintained. If the purpose of the study to examine relationships..then correlation analysis should be used. if the purpose or hypothesis intend to examine/compare between variables, then tests such as ANOVA or ttest should be used.
- Badges
- Science topic
- Similar topics
- Social Science
- Media
- Social Media