|
WHY DOES IT MATTER WHAT SORT OF DATA YOU HAVE? It comes down to deciding whether to use PARAMETRIC or NONPARAMETRIC statistical tests. The fact is, parametric tests are more powerful. If there is a real effect to be found, a parametric test will usually find it more efficiently than a non-parametric test - so it makes sense to use a parametric test if you possibly can.BUT: There are certain conditions about your data that normally have to be satisfied before you can use parametric statistics: 1) You must have data that are from a measure that is AT LEAST INTERVAL - nominal and ordinal are not good enough. 2) Your data must be from a population that has a NORMAL DISTRIBUTION. 3) If you are comparing samples, the variances within each sample must be similar - this is known as HOMOGENEITY OF VARIANCE. Each of these conditions is explained below. Scroll down or click on the highlighted words in the conditions to see these explanations. Type of data can be put in a sequence thus: Nominal, Ordinal, Interval, Ratio. As you progress from Nominal to Ratio, the level of measurement gets better in the sense that it gives you more and more information. Parametric statistics were developed for data that are at least interval level. So you can use them on ratio data too.NORMAL DISTRIBUTION - the bell-shaped curve This means that if you could plot data from the entire population on a histogram, it would look like a bell-shaped curve. That is, most cases would cluster around the mean value (average), and as you went out to either side, you would find progressively fewer cases, until, at extremely low or high values you have only a few cases. Statisticians have thoroughly investigated the properties or 'parameters' of this curve. PARAMETRIC statistics make the assumption that such parameters hold true. So if your sample is not from a normally distributed population, parametric statistics could give you wrong answers. NONPARAMETRIC statistics do not make this assumption. HOMOGENEITY OF VARIANCE This condition means that the different samples you are comparing must have similar variance - that is, the spread or dispersion of values must be about the same. Another measure of dispersion is the range - that is the distance from the smallest value to the largest. EXAMPLE: If you took a sample of people from the general population and a sample of people from an undergraduate psychology course, which sample would probably have the greater spread in ages? Yes, the sample from the general population would probably have a greater spread - and a greater variance - in ages than the sample from the undergraduate course, which tends to take people soon after they finish school. You might find that these two samples have significantly different variance in age - and therefore a parametric test such as the t-test (for comparing ages between the two samples) may not be appropriate. Back to Which Test Home Page |