Measurement System Analysis of Survey Data
- September 25, 2018 at 6:19 am #56100
I am conducting a study to understand if the users of a particular software are able to understand the codes in the software. I am conducting a sample survey consisting of 10 questions of which most of them are multiple choices which i can convert them in to a scale from 1-4. I would like to perform a MSA on the survey data. Is there any reliable method or examples that would help me to perform this analysis. Please help out.
JaySeptember 25, 2018 at 9:19 am #203062
Vijay You appear to be converting your data to a Likert Scale. You might want to do some research on what you do with Likert Scale data.October 1, 2018 at 4:30 am #203072
ETC tus información hoy can answer some hypotesis test. The MSA that you must to do is about to how you recive and record the information, because you are reciving perception of the people, and this is not repetibleOctober 2, 2018 at 2:10 pm #203086
Vijay, for perception based surveys I recommend evaluating the reliability by issuing the assessment twice and determining if there is a statistically significant correlation (p<.05) between responses using a chi-square test. That is to say, when I give a response of ‘3’ in the morning, I should also give a response of ‘3’ or ‘4’ in the afternoon when I take it again. If I don’t, then there’s likely questions that are ambiguous or double-barreled. You won’t analyze this like an MSA where you’re comparing variation between operators or parts in addition to the device…it’s all about the questions and how they’re written.
If you’re using statistical software like Minitab, simply title two columns as Test A and Test B; they’ll be the same survey but in Test A you’re going to list the first set of responses and in Test B, you’re going to list the second set of responses. Be sure that the list is in the same order by participant and question number. So if you start with Person A and question 1 under Test A, be sure that you start with Person A and question 1 under Test B and so on. Then run the chi-square. If you meet the above criteria (p<.05), you’re good to go but if not, keep reading.
I assume you would have put these in question order (i.e. 1, 2, 3…) so add a 3rd column titled ‘Question’ and enter the question number next to each entry; should be able to copy and paste after the first set. Now add a fourth column titled ‘Difference’ and put in a formula that subtracts the value of Test B from Test A from each entry. Then, assuming you’re using Minitab or some statistical software, run a graphical summary or whatever tool you have to determine the standard deviation of each question; I know this feels like a cardinal sin generating standard deviation using a Likert scale but stay with me. The question(s) with the highest standard deviations are the ones most affecting your chi-square result…the first response to the question is not correlating to the second response to the same question by the same person. Look at re-wording or replacing these questions with easier to understand/interpret questions. Once you’re satisfied with the re-wording/replacements issue the test twice again and repeat the chi-square test; you don’t have to issue the survey to the same audience since this isn’t an MSA. Good luck.
You must be logged in to reply to this topic.