Comparing multiple means from a single sample post hoc in SPSS

Question

For my research project I measured customer satisfaction on a 7 point Likert scale. For each business unit we measured satisfaction on 10 different items. On the first sight, the means seem to be pretty close to each other. I would now like to use a statistical procedure to see if any of those 10 means within a single business unit is significantly higher/ lower than others. Eventually I would like to rank the items (if possible) from high to low satisfaction.

 

In total there are 8 of these tables so by running paired T-tests there is a high chance that 5% will turn out to be significantly different just by chance, I would like to use a test that accounts for this problem. I also looked at ANOVA, but I do not have a dependent variable. All the data is already in SPSS, so it would nice if the analysis could be done with SPSS.

 

So in short: there is 1 sample group; there are 10 means ; there is NO independent variable ; method should account for error 1 ; preferably, the outcomes can be ranked.

Answer 1

Just to be clear then are you saying that the independent variable is which item the respondent completed? That is to say you are comparing scores to each other? 

 

If I am understanding that correctly then you may be interested in conducting a repeated measures ANOVA with something like a Tukey's post hoc test to identify specific differences. This would be appropriate if... 

 

1) Each data point comes from the same respondent/individual/business (whatever your unit of analysis is). That is the variables are being compared within subjects

 

2) Your outcome variable is continuous and approximately normally distributed.

 

The overall model will indicate whether there are differences between the 10 items and your post hoc tests will identify which differences are significantly different from zero. Tukey's is commonly used, but certain types of post hoc tests may be better/more appropriate depending on the nature of your data.  

 

To do all of this in SPSS go to Analyze -> general linear models -> repeated measures. 

 

You will create a factor with 10 levels (one for each of the items you are interested in), then you will click Define on the bottom left of the dialogue box and add the variables as within-subjects factors. There should also be an option to add post hoc tests. 

 

Because there will be a number of post hoc comparisons, it may be appropriate to set a more stringent alpha (referred to as a Bonferroni correction). You divide your initial alpha (.05) by the total number of comparisons being made and this reduces Type I error inflation. You will have to do the ranking yourself (which you can based on the means and based on which means differ significantly from one another).

 

I will say just one additional thing - that this is not a common way survey data is analyzed if the 10 items are intended to measure the same construct. If you could tell me a little more about the data and/or your research question, I might be able to provide a little more guidance or even suggest some statistical models that might better answer the kind of questions you are interested in answering.

 

Either way I hope this helps.