0
$\begingroup$

Whist reviewing some test data I am looking for advice on identifying the correct statistical tool(s) to help me answer two questions. The data I am looking at is from a cytotoxicity test, whereby a single categorical variable result is returned for one test. A "0" denotes no cytotoxic response, whilst a "4" denotes the highest possible cytotoxic response. The acceptance criteria is that the cytotoxicity must be equal to or less than 2. I have access to Minitab software for analysis.

I have two sets of data: before and after an event which caused a shift in cytotoxicity results. I want to answer two questions:

1.) Are the proportions of each category significantly different between the two data sets?

2.) What is the probability of producing a score >2 for each sample data set at some point in the future, assuming no other shifts in the process?

I believe I can answer the first question using the Chi-Square Goodness-of-Fit test. However, I feel like I'm clutching at straws for the second question. I am thinking about the binomial probability of success but this only works with attribute data (pass/fail) and not categorical.

A basic bar chart has been included below using some example data to illustrate my point. As you can see the population on the right has a "wider" distribution compared with the left. Assuming I had more data, can I produce a probability that I will receive a score of 3 or 4 at some point in the future?

Frequency of cytotoxicity scores by category

Improve this question
$\endgroup$

1 Answer 1

Reset to default
0
$\begingroup$

OK, I think you only need a simple test of proportions where you are testing that $H_0: p_0 = p_1$ where $p_i$ denotes the population proportion for cytotoxic. In Minitab I believe this is called a two-sample proportion test.

You have 0 cytotoxic responses before our of 133 trials, and 0 out of 133 after.

I think that is exactly as you describe.

If you want more information, I would fit a proportional odds logistic regression model with a single predictor variable (again, before/after), but that would let you model each level of your outcome variable.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Thanks for your suggestion Paul. You are correct, it is called exactly "2 Proportions" and I can see how that would be used. For the score of 3 you are again correct, the proportions are identical because both sets of data have 0 observations. However, from a logical stand point, if you were to collect an additional n=1,000 data points over time, you would naturally assume that the "after" data set is more likely to contain a 3 compared to the left data set. Is it possible to express this from a statistical point of view, or is it simply 0 because you have not observed it yet? $\endgroup$
    – IcedT
    Mar 5, 2020 at 15:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.