I need your help, as I am stuck with data I collected for a project. I am currently gathering responses about mitigation strategies against fatigue. I am questioning people about what mitigation strategies against fatigue they have experienced as helpful.
It is a multiple choice question that has 5 different strategies to choose from. More than one response is possible.
My research hypothesis is that one one of the strategies is perceived as more effective than the others. My original thought was to simply compare the counts of how often each strategy was chosen.
But how do I determine whether the difference is significant? What kind of test do I use?
How do I handle the data now?
I'd be super thankful if any of you could provide useful input. And if you do, please pretend like you are explaining to somebody who does not know much about stats ;-)
I take from what you written that each subject has rated each strategy with a yes/no response. These can be written 1 for yes and 0 for no. The data is assumed to in a rectangle with subjects going down and strategies going across.
One possibility is Cochran's Q test. This can be done in SPSS and there are tutorials on the net. Or you can do it in Excel and there is an example you can follow at http://www.real-statistics.com/anova...chrans-q-test/ or there is even an online calculator at ttp://scistatcalc.blogspot.co.nz/2013/12/cochrans-q-test-calculator.html
Thank you so much for your response. Sorry for not being clear, but the respondents do not choose a yes/no for each strategy. Rather the questions goes like this:
What mitigation strategy or strategies have you personally experienced as helpful? (multiple answers are possible)
That leaves me with a count how often each strategy was clicked on and the according percentage. I now want to determine whether one of them was clicked significantly more often than others. From what I understand when reading into your provided links (THANK YOU FOR THAT!!), the Cochran's Q test is not the right fit. But I might be wrong....
What do you think given the different understanding of how the question was posed?
So, each person can click as many as they want to - is that right?
Do you have the various responses for each person, or only the total number of clicks at the end? In other words do you have Person 1, A. Person 2, B and C, Person 3 none ...? or do you have only A 25 people, B 13 people, C 0 people ... out of 125 people?
I have both a summary and I can also look into each person's response individually.
I was hoping that the ' A 25 people, B 13 people, C 0 people ... out of 125 people?' option would work for simplicity reasons ;-)
You can certainly just use the summary data to find the percentage of each one. You can also find the uncertainty in each calculated percentage.
There are also comparatively simple ways of finding if two or more percentages are different if they use independent data - the Chi square test for example. However, in your case, the data isn't independent because each person gives several responses.
So, it depends how strictly correct you want to be and what is expected from your project. You do have the data to do Cochran's Q test, but this may be overkill unless the report is going to be peer reviewed. A simple alternative could be to work with each strategy separately, and find each proportion and its margin of error. Then draw a graph of the proportion for each strategy with error bars and let people decide for themselves. But, if you want to quote honest p values, and be protected from criticism from some reviewer then you need an analysis which takes into account that each person contributes several times to the data (repeated measures).