04-24-2017 03:03 PM - edited 04-24-2017 03:11 PM
I am working on a similar kind of project as this
For table 2, what are the appropriate statistical tests to use? (we have only the count of cataract surgeries by age group, gender and race)
Thanks in advance,
04-24-2017 03:16 PM
The note on that table reads, "Significance level from Student's t-test (gender comparison) or Analysis of Variance (age and race comparisons)." So if you are trying to replicate their results or compare results from your experiments to theirs, that would be a good start.
I recommend the book Statistics in Plain English, or whatever intro statistics textbook you may have used in your stats class, for revisiting the logic behind these tests, and where they should be used. It's far better to understand the basic stats behind the tests than to simply select one and throw data at it.
04-24-2017 03:19 PM
Thanks for your reply.
But since we are using the counts, i thought Chi-suare would be appropriate. Not sure how we can use Student's t-test (gender comparison) or Analysis of Variance (age and race comparisons) for counts!
04-24-2017 03:30 PM
You would use chi-square to compare two categorical variables. However, I would recommend logistic regression as you could get the p-values and confidence limits for each effect.
04-24-2017 07:56 PM - edited 04-24-2017 07:57 PM
I am actually working on a similar study. I thought of using Chi-square for p values. I did some research on similar kind of studies and found this, however, did not understand how to use Student's t-test (gender comparison) or Analysis of Variance (age and race comparisons).
I was wondering if chi-square is appropriate or not?
04-24-2017 08:09 PM
My guess is if they did t-tests or ANOVA it was on the rates, not counts. Validity of that is debatable but the numbers are large enough to not cause too much concern on my part.
One thing, NEVER assume that the analysis used is appropriate and what you should be doing, make the decision based on your data and statistical knowledge, you have to be able to defend it and there's too many studies that demonstrate statistical issues these days.