I am working on research involving the collection of data from employees and customers. Specifically, my research question deals with the influence of employee satisfaction on the quality of customer service. I will collect data from both employees and customers, and plan on running a correlation analysis on the results. I know i will run canonical correlation, but my question is, if i have collected 200 questionnaires from employees, and I have 400 from customers, in a single SPSS file, there is an issue of missing data, since employees data is 200 less than the customers data. Does anyone have a suggestion for how I should handle this?
$\begingroup$
$\endgroup$
1
-
$\begingroup$ Different number of variables in the 2 sets is not an issue, it is normal. If you have missing data within the sets, try to imput them, e.g. via EM algorithm. $\endgroup$– ttnphnsCommented Dec 31, 2012 at 8:20
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
Canonical correlation is not your tool, but possibly you don't mean that method because you also refer to 'correlation analysis' which is different.
In the simplest case you have an instrument that measures satisfaction, e.g. a question in a survey, and you have 200 measures from employees and 400 from the customers. You don't have missing data unless the answers are coupled somehow. As @F.Tusell suggests, if the survey was given to the customer who complained and also to the employee that dealt with the complaint then they would be analysed together. As you describe it you just have data in two groups with different sizes.
In the absence of anything coupling the two sets of responses you might be interested in difference between these two satisfaction distributions. One very simple but informative question might be: do customer and employee satisfaction levels differ on average? This is the sort of thing answered by a t-test.
answered Jan 30, 2013 at 11:53
$\begingroup$
$\endgroup$
2
I do not quite understand your setting. Canonical correlation will let you examine (linear) relationships among sets of variables, but they have to be paired. For instance, the employee satisfaction variables and the customer responses on quality service would correspond to a pair employee-customer which have interacted. It doesn't make sense to correlate data on employees and customers which have never interacted.
With that in mind, your 200 employee observations and 400 customer observations might be the outcome of a single employee serving (on average) two customers. I think you could pair each customer record with the record of the employee which provided the service, and thus have 400 full records: no imputation required, only pairing.
-
$\begingroup$ Let me Reask Sir. My Research topic is Impact of Internal Service Quality on External Service Quality, the data will be collected from Employees for Internal Service Quality and for External Service Quality the Data will be collected from Customers. Now i want to correlate the data. Since 2 Questionnaires are filled by different set of respondents, the number of questionnaire filled will definitely change. How do i see if there is a relationship between internal service quality and external service quality. $\endgroup$– FawadCommented Dec 31, 2012 at 9:52
-
2$\begingroup$ I do not think you can correlate data from two different surveys. Canonical correlation analysis is not the right tool for that, nor do I think there is any other. $\endgroup$ Commented Dec 31, 2012 at 11:33