Hello,
I have little doubts about the statistical model that I am using to analyze my data. I have two groups of residue studies data Group 1 n=7 and the other group 2 n=47. they are independent and the studies are expensive and rare so I couldn't increase the sample size by any means.
I have tested the normality of both groups using the SPSS and found to be not normally distributed then I transformed all the data to fit the normal distribution using the square-root calculation - they fit the normal distribution p= 0.2 (more than 0.05).
the data is then analyzed by SPSS - independent t-test to compare between the two means and the p-value of unequal variance was selected to act as the significant value instead of the usual T-test (by other words Welch's test was used ) kindly have a look at the attached data
and used to plot the CI of 95% with each mean of the groups tested in a bar chart
Do you agree using a sample of 7 is not affecting or biasing the test result? and if n=5 of one group and the other group n = 16 is is still a valid method to use? or should I use another test for small samples?
I wish if someone confirm my thoughts?
Thanks
Use the freely available JASP statistical program so you can strengthen your results with Bayesian Tests.
Article Bayesian inference for psychology. Part II: Example applicat...
Sure, as long as you can estimate a standard error. It won't be powerful though so depending on the context might not be useful. With only 7 and such a low standard deviation I'd be careful with the comparison, but you can put the numbers and an answer will come. So it is possible. Whether you should is a completely different question.
Daniel Wright
Thank you so much for the answer
if I understood you well , you mean that analysis using T test is posisble with small sample size . But in may case I should be more careful because I have low standard error ? I though small stand error is advantageous ?
do you recommend doing any thing rather than this test? As from this conclusion of P ( very low) I can say there is a significant difference between the means of the two groups
do you agree Daniel ?
Thanks
With small numbers people often use Wilcoxon ranked sum test, but it depends on your research questions. If you have a question about what you should do, not what is mathematically possible, I recommend starting a new question. You will get more responses.
I think you have already got your ans from Dr. Daniel Wright. Yes, you can but the bigger sample with more number of different groups will definitely help you to get more validated and representative answer.
Thank you Muhammed Ashraful Alam
unfortunately I cant increase the sample size by any way as this is the only data available
thanks for confirming the answer
Hello Mohamed Badawy. So far, no one has commented on your use of normality tests prior to doing a t-test. Generally speaking, using statistical tests of normality or homogeneity of variance to determine whether one can safely use a t-test (or ANOVA, etc.) is a bad idea. I gave a short conference presentation on this a few years ago that you might find interesting.
Presentation Silly or Pointless Things People Do When Analyzing Data: 1. ...
I don't know what residue studies are, so do not understand the nature of your dependent variable. But your use of a square root transformation made me wonder if it is a count. Even if it is not a count, here is an article I found just now that looks relevant. It suggests that you might consider a GzLM with a Poisson error distribution.
Article Count data in biology-Data transformation or model reformation?
HTH.
Bruce Weaver . thank you so much for the information.It seems professional and I will need to have a detailed look.
I means by residue studies : pesticide residue on and in plant matrices compared to residue on plant leaves ; they are extremely independent to each others and know to be not normally distributed due to many different physiological and metrological conditions associated with each study .
Thank you again
- Badges
- Science topic