Hi,
I am doing my research about the influence of managers on staff sustainable behaviours. I would like to correlated that when staff perceive the support from managers for sustainability they are likely to perform sustainable behaviours. My sample is now 25. Can I use regression with this size of sample. If not, is there any suggestion how can I achieve it?
Thank you very much,
Nhat
* the sustainable behaviours is measures on the scale of very frequently,frequently, occasionally, rarely and never.
And the perception on managers support is on strongly agree, agree....strongly disagree scale.
I think you need a slightly bigger sample 25 maybe a bit too low. U can make use of G-Power analysis or any other technique to determine sample size to help you decide on a sample size that suit your cost, objectives at the same time giving you reasonable accuracy.
Hello, I agree with the others as I also think you should increase your sample size. It also depends on how homogeneous is your sample (size and industry of companies, age of the managers, country where they operate), the more heterogeneous the more cases you need. However, there are several other limitations with your approach that are not going to be solved by getting more data, namely: lack of control variables and, as far as I understand, use of two single item measures on Likert-scale. It's going difficult to establish causality anyway... Given also the newness of the topic, you could consider performing some in-depth case studies.
I don't agree to the others.
1) It always depends on the size of the effect you aim to see and the risk of a "false-negative" result one is willing to accept. I don't believe in rules-of-thumb - you always have to have any opinion about a relevant effect size and acceptable risks to get a "false-positive" or a "false negative" result (alternatively, you have to have an opinion about the desired precision of your effect estimates). But, well, maybe I would accept one "rule-of-thumb", which is: in basic research, always use the largest sample size that is feasable (withing the given limits of time, money, workload). That's it. If you then can see something in your data, fine. If not, then the topic is not feasably to be worked on anyway.
2) The response is ordinal. I assume that with "regression" is meant the "standard" linear regression - what is inappropriate for such data. If some regression model is used, then it should be an ordered logistic regression.
You can always try to run a test analysis of your data. Your regression results will give you a p value that will give you some indication of whether you are close to a usable sample size. If your p is
Roscoe (1975), cited by Fisher (2007), advise as a rule-of-thumb, a sample size of 30 is acceptable for statistical analysis. But, if the sample size is very small, only very strong relationships will be demonstrable.
If you have more than one independent variable, as much as possible use at least 20 observations for each independent variable to be studied. Say, you have 3 independent variables in your study, you have at least 60 observations as your sample.
BUT, very much agree with Sir Pedro, that MORE data is always BETTER.
Whatever you do sample size of 25 limits highly generalization of your results to a wider population. I would strongly suggest to increase your sample size unless it is an exercise. Guidance on that vary you could use G power application that is free online or follow Ertugrul suggestion.
Dear Nhat
The sample size for any statistical analysis depend on standard deviation among observations and the margin of error for your test , you can use many softwares to estimate appropriate sample size as SPSS , G*Power 3.1
Good Luck
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression