# Nonparametric test and ANCOVA

#### Tiing

##### New Member
Good day everyone.
I'm a student who are currently doing her psychology research (final year project). I'm having an issue with my data being not normal. I have ran the test of normality(Shapiro Wilks test) using SPSS software and I found out that my p-value is 0.000 (not significant). And then I ran independent factorial ANCOVA and it turn out that my 2nd IV (Main effect: types of task) is significant only.
Recently, I have read through the statistics book and I found out that if the data is not normally distributed then need to run non-parametric test or do data transformation or remove outlier. Nevertheless, my lecturer always ask us to assume the data is normal in every single class project because we are only taught about parametric test but I'm currently doing final project so I can't ignore my data since it is not normal. That's why I'm very confused which option should I go for. I'm very shocked when I read about it (if the data is not normally distributed then need to run non-parametric test or do data transformation or remove outlier) because I have no exposure about non-parametric test or do data transformation or remove outlier.
Please lend a helping hand and your help is much appreciated. I'm very weak in statistics because I just barely pass this subject. I have tried very hard to read and understand the statistics which not taught in class. I hope that you can give my some advice because we are not allow to ask any lecturer about the statistics.

The following is some brief information regarding my research project:

There were two independent variables (IV) in this study. The first IV was gender, which contained of two levels, namely female and male. The second IV was types of task, which also have two levels, specifically single task versus multitask. The dependent variable (DV) was a mathematical task performance. The operational definition of the DV was the total number of correct responses generated out of 12 mathematical questions within 10 minutes.

Sample size required (based on G power): 125

Remaining usable data in my study:122

The normality was assumed for: female on mathematical task performance (skewness = -1.25, kurtosis = 0.77), male on mathematical task performance (skewness = -0.80, kurtosis = -0.75) and multitask on mathematical task performance (skewness = -0.20, kurtosis = -1.12) because the skew and kurtosis fell within the range of -2 to +2 .The normality was not assumed for single task on mathematical task performance (skewness = -2.73, kurtosis = 8.55). Thus, a negatively skewed distribution was displayed for: female on mathematical task performance, male on mathematical task performance; multitask on mathematical task performance and single task on mathematical task performance.
Based on the Shapiro Wilks W test method, the data was not assumed to be normally distributed for: female on mathematical task performance, male on mathematical task performance; multitask on mathematical task performance and single task on mathematical task performance because the Shapiro-Wilks p-value was less than .05.

Thank you very much.

Tiing

#### Karabiner

##### TS Contributor
Not the dependent variable should be normally distributed,
but the residuals of your prediction model. And in case of
a sufficiently large sample (> 100 is often considered sufficient)
non-normality is usually not invalidating the statistical
significance test (if there are not influential outliers in the
data).

With kind regards

K.