|
The answers to the final three questions determine which test you will actually use. For instance, if you are satisfied that your data fulfill the requirements for a parametric test, and you have manipulated more than one IV, then you will probably look for the appropriate Analysis of Variance (ANOVA) to perform. (See Section 4.6.10 of the book for more about how to use and report ANOVA). On the other hand, if you only wish to compare two conditions on one IV, you are not happy to make the assumptions necessary for the parametric test, and you are comparing data from the same participants, then you might consider using the Wilcoxon signed-ranks test. (See Section 4.6 of the book and Section B of this Web site for more on when to use the commoner inferential tests and how to report them.)
You can see from question six that the type of test that you use will also depend on whether or not you are making comparisons between unrelated or related samples. (If you are not familiar with the terms unrelated and related samples, you will probably have been taught some different terms for the same thing. You can find out which terms mean the same thing as unrelated and related samples in Section 10.2 of the book.) Thus the issues discussed in Chapter 10 of the book directly affect the type of statistical operations you will perform on the data. However, there is one potential source of confusion here - whether or not you should consider data from studies in which you have matched your participants (Section 10.7 of the book) to be unrelated or related for the purposes of analysis. For more on this see Section A4 of this Web site.
Another common source of confusion here stems from the feeling that there should be one best way of analysing a given set of data. However, it is not at all unusual to find that there may be more than one way of meaningfully analysing a given set of data. For instance, even where your data fully satisfy the requirements of parametric tests, it is still perfectly permissible to analyse them with an appropriate nonparametric test. (The reverse, however, is certainly not true - that is, there are data that should only be analysed nonparametrically.) Moreover, under these circumstances you might even find that you reach a different decision over whether to reject or not reject the null hypothesis. This is because parametric tests are generally more powerful than their nonparametric equivalents, and power in this sense refers to the probability of correctly rejecting the null hypothesis (i.e., of rejecting the null hypothesis when it is indeed false). You can find a discussion of power in Chapter 12 of the book.
For the purposes of your report, however, you must never duplicate tests in this way. That is, under these circumstances you should choose to perform and report only one of the relevant tests. In the book I highlight common mistakes that students persist in making despite my advice to the contrary, and this is one such mistake!
Once you are familiar with the concepts described here, you might like to try your hand at the following SAQ:
The studies below are those described in SAQ24 of the book. For full details, see SAQ24. Go through each in turn, assessing for each the answer to the above questions and arriving at a recommendation for a suitable statistical test.
How might the researcher find out whether there was any effect of:
- word frequency on time taken to decide whether a stimulus was a word or a nonword (in milliseconds)?
- oestrogen on body weight (in grams)?
- the level of anxiety on the number of viewers who took up the opportunity to make dental appointments?
- the nature of television violence on the mean level of shock (in volts) administered by the viewers of the violence?
- working with others on the number of packets of breakfast cereal packed in 20 minutes?
Click here for the answer |
