# What tests to use for analyzing data that is categorical and continuous

#### MVW97

##### New Member
Hey everyone,

for my master thesis, I'm analyzing a bunch of patient data. I have very limited statistical experience hence the question here.
The data set includes both categorical variables and continuous variables. the continuous variables don't seem to be normally distributed. additionally, these measures from the continuous variables are from a patient group and there is no data available on those specific variables in a healthy control group.
What would be the best way to analyze this data? the goal is to compare and correlate all variables. But would it be best to first transform the continuous variables into categorical data? If so how to do this without having normalized data? Or could I use the categorical variables to predict the continuous data?
Many thanks for the help and i can give more info if needed!

Myrthe

#### sl8

##### New Member
Hello. Based on the description of data:
1) Association between a categorical variable and a quantitative variables:
If the quantitative variables are not certain to be normally distributed, then I suggest to use the following non-parametric tests:
1A) Mann-Whitney U test to compare the quantitative variable between two categories of a categorical variable.
1B) If a categorical variable contains three or more categories use the Kruskal-Wallis H test.
2) Association between two categorical variables: three basic principles need to be considered before applying the right test:
A - The total sample size of the compared categories, B - the "expected" values of each cell in the table of the compared categories, C - independent or paired samples.
If the total sample size is above 20, then use Pearson's Chi-Square test. If not, if the sample size is less than 20 then use the Yate's correction for continuity.
Concerning the "expected" values, if at least one cell has an expected value of less than 5 (<5), then it is recommended to use the Fisher's exact test for association between dichotomized variables (2x2 table), or the Fisher-Freeman-Halton exact test for a 2x3 table.
If the samples are paired, then use yhe McNemar test for a 2x2 table.