PDA

View Full Version : SPSS - Missing items on a scale

Staticia
05-28-2008, 06:06 AM
I am working on a turn paper and was provided with a set of data. This includes a scale consisting of 7 questions that all have 4 different answer categories. I have given the categories the numbers 0, 1, 2, 3, and combined them in a new variable where the numbers are summed (0 being the lowest possible score and 21 the highest).
The problem is I want to include everyone that has responded to at least 6 of the 7 questions, no matter which one they did not respond to. Does anyone know how to do this? Right now everyone that is included in my new variable have responded to all 7 questions, making the number of cases not included too large.

Savvy
05-28-2008, 06:50 AM
Enter all your data, just leaving blank spaces where you have missing data. When you go to analyse, under options it will ask if you want to exclude cases listwise or pairwise. You would want to choose pairwise.

Staticia
05-28-2008, 06:27 PM
Thanks for your response! I have now tried to understand the "pairwise" option, but being new to SPSS and statistics I am still a little confused. For instance, the help function in SPSS has this explanation:

"Pairwise: Delete cases with missing values pairwise. A case is deleted from the analysis only if it has a missing value for the dependent variable or factor being analyzed".

In my case the total scale is the dependent variable. How will SPSS "know" to include cases that have a missing value on one of the subscores, but not cases that have more than one?

Say one of my cases have this response to the subscores:

#1 - 3
#2 - 2
#3 - 1
#4 - 1
#5 - missing
#6 - 1
#7 - 2

That would give a total score of 10, plus I need to include the missing value. I think i will do this by assigning it the average of the other scores (10 divided by 7) and add that to the 10 points. The sum of this would be this case's value on my dependent variable.

I want to include this case in my analysis, but if they had one more missing value on the subscore I would not. How does the pairwise exclusion deal with this?

I appologize for my ignorance, and any attempt to sort out my confusion will be highly appreciated.

Savvy
05-28-2008, 08:21 PM
If, for example, you are running a corrrelation, and you have missing data from a participant on Q.5, a pairwise exclusion will include the participant's entire set of data in the correlation EXCEPT Q.5. In a listwise exclusion, the participant's entire case would be dropped, leaving you with too small a number of participants. So SPSS really takes care of it for you.

The other thing you can do is mean substitution. IE, you can average the scores of the rest of the cases on that particular item, and give that participant that score. This however, will reduce variance.

There is also on SPSS the option of expectation maximisation: Go to analyse> missing value analysis> EM. But you have to ensure normality of distribution to use this option. This doesn't effect variance as much as mean substitution.

Hope this helps.