The Student Room Group

Which statistical test should I use?

If I wanted to find a relationship between handspan (in mm) and earlobe attachment (where 0 is detached, 1 is partially, 2 is attached). What test should I use? I was considering a Mann Whitney but I don't know if this is correct
Reply 1
Original post by no-brainer
If I wanted to find a relationship between handspan (in mm) and earlobe attachment (where 0 is detached, 1 is partially, 2 is attached). What test should I use? I was considering a Mann Whitney but I don't know if this is correct

Firstly, you need to decide whether you're going to use a parametric or non-parametric test. Parametric tests make the assumption that the data is normally distributed whereas non-parametric tests do not. Handspan is likely to be normally distributed, similar to other variables like height and weight. The best way to check whether the data is normally distributed is by simply drawing and inspecting a handspan against frequency plot. There are statistically methods to assess whether your data is normally distributed including the Shapiro-Wilk test (if you have less than 50 measurements) or the Kolmogorov-Smirnov test (if n>50). If in doubt, use a non-parametric test. For each parametric test, there are non-parametric equivalent tests.

Once you have decided whether your data is normally distributed or not (and therefore whether you are using a parametric or non-parametric test), you then need to look at how many variables you have. You are comparing the handspan across three groups, therefore you should use the ANOVA test (parametric test i.e. if data is normally distributed) or the Kruskal-Wallis test (non-parametric i,e, if data is not normal). However, these tests will not tell you which group(s) are statistically different from the others, they will just tell you that at least one variable is significantly different from another.

The Mann-Whitney U test is a non-parametric test for comparing the medians across two groups; and the parametric equivalent of the Mann-Whitney U test is a 2-sample t test. Therefore, a Mann-Whitney test cannot be used here as you have more than two groups.

Hope that helps.
This is what I want to recommend. Choose between a non parameter test or a parameter test. Once you have selected one then decide on variables etc. If you are not sure I would strongly recommend a non parameter test. But the specific test in question is not recommended as you have more than two groups. These tests have their limitations and risks too.
Reply 3
Original post by Jpw1097
Firstly, you need to decide whether you're going to use a parametric or non-parametric test. Parametric tests make the assumption that the data is normally distributed whereas non-parametric tests do not. Handspan is likely to be normally distributed, similar to other variables like height and weight. The best way to check whether the data is normally distributed is by simply drawing and inspecting a handspan against frequency plot. There are statistically methods to assess whether your data is normally distributed including the Shapiro-Wilk test (if you have less than 50 measurements) or the Kolmogorov-Smirnov test (if n>50). If in doubt, use a non-parametric test. For each parametric test, there are non-parametric equivalent tests.

Once you have decided whether your data is normally distributed or not (and therefore whether you are using a parametric or non-parametric test), you then need to look at how many variables you have. You are comparing the handspan across three groups, therefore you should use the ANOVA test (parametric test i.e. if data is normally distributed) or the Kruskal-Wallis test (non-parametric i,e, if data is not normal). However, these tests will not tell you which group(s) are statistically different from the others, they will just tell you that at least one variable is significantly different from another.

The Mann-Whitney U test is a non-parametric test for comparing the medians across two groups; and the parametric equivalent of the Mann-Whitney U test is a 2-sample t test. Therefore, a Mann-Whitney test cannot be used here as you have more than two groups.

Hope that helps.

Thank you so so much!!
Reply 4
Original post by tinygirl96
This is what I want to recommend. Choose between a non parameter test or a parameter test. Once you have selected one then decide on variables etc. If you are not sure I would strongly recommend a non parameter test. But the specific test in question is not recommended as you have more than two groups. These tests have their limitations and risks too.

Thank you!!

Quick Reply

Latest