The survey includes questions asking about the general quality, duration, instructions, sufficiency of the activities provided, feeling confident to evaluate the mental health condition presented, feeling confident to manage the condition presented. This survey was given to the participants after each of the 8 modules ( on different mental health disorders) presented. Each question has 3-4 options, following the Likert scale in some cases. For general quality, the possible responses are: Very good, adequate, needs improvement. For questions asking about the participant's confidence level though, possible responses are: Yes, Partially, No.
Currently I have categorized all the responses for each question (from each module) into one bar chart, depicting the frequencies. However, I would also like to be able to say that the general quality of the training was __% positive. Or that for a certain question, the proportion that responded positively was statistically significant, compared to those who responded negatively, with a p value.
Any suggestions on which statistical tests to use, with Excel. Suggestions and tips would be truly appreciated!!!
Sample size is 33, but for some questions, no more than 3 responses are missing.
Put everything in a table first, then use non-parametric statistics for small samples.
You can directly calculate all the proportions in excel. and chi square can be used to check the difference in proportions of positive and negative responses with a p value
Yeah, go the the non-parametric chi-square that you will know whether the proportion who say yes for example is different from the proportion from those who said no.
I think you have answered the primary part of the question yourself. Probably the most substantive thing is to calculate the percentage of each response for each question. Or, as you say, you can compress the responses into "positive" and "negative" in order to report the percent positive responses. Doing this, you are reporting the results of each question separate from the other questions.
If you want a p-value for the percent positive responses for a single question, the correct test is the binomial test. For this test, you have to have a proportion to compare the measured proportion to. In this case you would probably just use 0.5. So, you are testing if the percent positive responses are statistically different from 0.5. You are calculating this for each questions eparate from the other questions.
Look up the binomial test in the following two links, or elsewhere. You want a two-sided test following the Gus the Cat example. You can probably do this test in Excel. You can use the examples here to make sure that you are doing them correctly.
http://www.biostathandbook.com/exactgof.html
http://rcompanion.org/rcompanion/b_01.html
This calculator appears to work fine:
http://graphpad.com/quickcalcs/binomial1.cfm
I don't think using a chi-square test will answer the questions you are asking.
One more note. If you are testing the hypothesis "The percent positive responses for this question was statistically greater than 0.5" rather than "...statistically different..." , you can use the one-sided test (one-tailed test) for the binomial test.
Thank you so much everyone for taking the time to answer and provide such great feedback! I am going to take these into consideration.
Well I am a bit conflicted now. Considering that I have at least 3 options for responses to each question ( and they are difficult to compress into positive and negative; these are 3 distinct categories ) - for e.g. Are there sufficient participatory activities? Responses are: need more, sufficient, need less. This changes the situation for which test I can use now...
It does not change anything but to give a good answer you would like to know your research question. In case you just explore your data construct a big table with frequencies and possible differences between men and women. There is an old good book for non-parametric statistics where all the conditions and test suggestions are given inside the cover page: Nonparametric Statistics for Behavioural Science 1956
by Sidney Siegel (good)
Newer edition: Nonparametric Statistics for The Behavioral Sciences Jan 1, 1988
by Sidney Siegel and N. John Castellan Jr.
Okay, thanks a ton, Beatrice! I have all the frequencies depicted visually, in graphs, at the moment. I am ultimately trying to evaluate the training from all these questions (and other qualitative ones that were also a part of the survey) in order to improve it, or keep it the same for future training. I will look into finding that source or similar online.
You can compute a variable for the general training by summing the scores of each modules ( analyze in SPSS and the compute variable) and then use the sum, which has a new name as a dependent variable in a linear regression analysis, then you look at the part correlation if it is significant. If it is significant, you square it and then you have the % you are looking for. If the part correlation is , e.g. .430, you square it and get 18% explained.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size