I planned to run independent t-test on my data; however, my data were not normally distributed with outliers retained in the dataset. I have two dependent variables and one categorical variable. I removed the outliers and, still, the datasets were not normally distributed. I performed a Log10 transformation on the dataset with the outliers removed and this resulted in both dependent variable being not normally distributed for each category of the independent variable. Essentially, the transformation proved the non-normality of the dataset. With the data not being normally distributed, this prompted the use of the Mann-Whitney U-Test. So, do I run the MW on the transformed data or the original dataset? If I use the original dataset, do I include the outlier or not? The outliers were within the parameters of my instruments.
What statistical hypothesis do you want to test? Note that t-test and MW-test test different statistical hypotheses.
What is the response variable we are talking about?
What is the sample size?
How did you define "outliers"? Is there any argument other than that they don't "look good" in your assumed distribution model, that these values are untrustworthy? And of how many outliers are we talking?
Could you add a plot of the data against the categorical variable?
I should clarify that I am a doctoral candidate and not a statistician. I am comparing two dependent variables (the variable labels are TSE & TechInt) based on location categories (rural and non-rural). The outliers were identified via boxplots when I ran the descriptives in SPSS. My sample size is N = 154 (75 rural; 79 non-rural). For TSE there were 10 outliers with 5 each from rural and non-rural. For TechInt there were 3 outliers all in the non-rural category.
Apologies for the off-topic post, but John's first sentence above reminded me immediately of Dr. McCoy from Star Trek. ;-)
https://memory-alpha.fandom.com/wiki/I%27m_a_doctor,_not_a...
EDIT: Attaching an image. ;-)
- Badges
- Science topic