I'm hoping someone out there can help me with this SPSS problem. I'm trying to calculate an index score. However, the index score isn't a simple summation of several items - perhaps weighting some, perhaps recoding some. No. Because I know how to do that.
The scale is calculated as the average of the percentage of respondents that answer 4 or 5 on one item, a 4 or 5 on item 2, and a 4 or 5 on item 3.
Let me give you an example. Let's say we have 10 respondents. On item one, 5 people give a score of 4 or 5 - or 50%. On item two, 4 people rate a 4 or 5 - or 40%. On item three, only 2 people give a rating of 4 or 5 - or 20%.
The scale score would be (50% + 40% + 20%)/3.
Now yes, I fully acknowledge I could do this formula in excel. However, I'm not any good with excel beyond basic arithmetic. And, I've already created some new variables in SPSS and the file won't export into excel.
Now, I also realize I could do this by hand. Herein lies the problem. While the score is interesting, I want to be able to conduct t-tests for the people that responded to the measure before an event (pre) and after an event (post). The theory is that people after the event responded more positively or negatively than those who responded before the event. And, I further wanted to do an analysis of responses by day of the week to see if there was any pattern about when in the week was the most positive or negative results if at all (I don't think there is but my boss keeps saying it's true and we've never conducted actual analysis to verify).
As far as I've gotten at this point is it feels like dummy coding is needed. So 4 or 5s are now 1 and 1, 2, 3 are now 0. When I run frequencies I can get percentages...but now I am stuck.
Help!
I like to cheat and just copy and paste a couple of columns into Excel, do the calculation, and then (making sure the SPSS file hasn't been reordered), paste back in SPSS. I'm sure there's a way of doing it in SPSS, I'd have to play with it for a while to figure out.
ReplyDeleteFor the t-tests, I'm not sure that you can use your scale score...unless you're doing an independent t-test - since the 20% before may not be the same 20% afterwards and, therefore, the people wouldn't be matched. Response by day of the week should be straightforward, sounds like a one-way anova with 7 groups (assuming you're using all days of the week) and the number of out 100 as the outcome variable. Maybe a mixed design if you're using pre and post event measures.
The problems is that your defnition of 'scale score' is for the entire group not for an individual.
ReplyDeleteUnless you have worded your research question incorrectly--a simple independent t-test of means will answer your question of "were the people before event X more positive than people after event x".
Hope this helps! -statgirrl
You are both correct in the need to do an independent t-test as the hypothesis is that the people that responded before the event are more/less positive than those who responded after the event. Keeping it in SPSS, I couldn't come up with a way to calculate an individual score of positivity. Best approach was to dummy code each item and create a variable that summed the three items. A person could have a sum of 0 to 3. I also dummy coded pre/post based on the date of response. I played around with either using the pre/post as a variable or splitting the file on the pre/post.
ReplyDeleteIn the end, I took the advice to move into excel sort of. I split the SPSS file on pre/post and ran frequencies including the sum. In excel I entered the sum divided by 3x the n responses. I ended up getting pre/post scores as well as taking the same general approach for the days of the week. Eyeballing it, I wouldn't say there's a difference pre/post as there's a difference of 3%. However the post scores are based on a much smaller n, of which a few are suspiciously higher in some regards (I'd almost hypothesize that people who will be effected by the results either went in the last few days or had someone on their behalf to rate the items very high in order to try to pull the overall average up). Taking this approach, I'm not able to actually conduct any statistical test of significant difference.
An alternative hypothesis to the intentional skewing is if people that respond to survey reminders that this is the last day to do it are the ones that are generally more positive about the topic and didn't feel it necessary to respond but are being pressured into responding.
p.s, the score was made up of three items which is why I multiplied by 3.
ReplyDelete