Feedback for my correlations analysis plan

General
1

correlation analyses

Hello,
I would appreciate some feedback on my statistical analysis plan. I am a psychology PhD student and conducted an online study; I am currently performing the analyses, starting with correlations. My study consisted of two parts.
Total n (Part 1) = 818
Total n (Part 2) = 555
Given the large number of independent variables (IVs), I plan to use the Benjamini-Hochberg (BH) procedure to control the False Discovery Rate (FDR).

A. Correlation between a binary dependent variable (DV) and a continuous IV + significance test
For DV = 0: n (Part 1) = 413, n (Part 2) = 276
For DV = 1: n (Part 1) = 405, n (Part 2) = 279

Point-biserial correlation if: no outliers for the continuous variable within each category of the dichotomous variable; continuous variable is approximately normally distributed within each category of the dichotomous variable; continuous variable has equal variances across categories of the dichotomous variable.

If assumptions are not met: rank-biserial correlation

  • for each test: Cook's distance to assess whether a data point is influencing the correlation
  • FDR applied to the set of results

B. Correlation between a categorical DV and a continuous IV + significance test
n (Part 1) = 405, n (Part 2) = 279

Polyserial correlation

+for each test: Cook's distance to assess whether a data point influences the correlation
+FDR applied to the set of results

C. Correlation between a continuous DV and a continuous IV + significance test
n part 1 = 405, n part 2 = 279

Pearson correlation if the IV meets the assumption
Spearman correlation if the IV does not meet the assumption

+for each test: Cook's distance to assess whether a data point influences the correlation
+FDR applied to the set of results

My questions:
Does this seem correct to you? I have a doubt regarding Part B. How do I check for a correlation between a variable with more than two categories and a continuous variable?
My supervisor mentioned Cook's distance for assessing outliers. I'm not sure if it's useful for that purpose. Can it be used in isolation, independently of a model?

I have a doubt regarding Part B. How do I check for a correlation between a variable with more than two categories and a continuous variable?

My supervisor mentioned Cook's distance for assessing outliers. I'm not sure if it's useful for that purpose. Can it be used in isolation, independently of a model?

2

Hi and welcome to the forum.

I think your questions are too specific to answer without some larger view of the study design, some idea of what the data look like and why you want to use correlations in the analysis. Are you planning further analyses?

Can you give us an outline of the study—if you have an abstract that might do—plus some data. The actual data would be ideal but if it is confidential then simulated data should help, as long as it matches th rleal dada in characteristics, that is Var XX in the simulated data is character if XX is character in the real data, etc.

The best way to supply data is to use the dput() function. Do dput(mydata) where "mydata" is the name of your dataset. Paste the output here between

```

```
You may also find this helpful.

The RStudio Forum is mainly for R questions rather than data analysis questions. There are some very knowledgeable statistians—not me!— here but you might find Cross Validated (https://stats.stackexchange.com) a better place to ask questions.

If your university has a statistical consulting service, booking an hour with a statistician may be a good idea.

Why are you so concerned about outliers? BTW, probably the best thing to do with the data is graph them before seriously worrying about Cook's D.