How to calculate an overall job satisfaction score and test group differences?

Question

I am doing a project on finding out whether there is a difference in job satisfaction between employees of non-profit organisations and for-profit organisations. I also need to find whether job satisfaction varies as a function of age (categorized as below 20, 21-30 and so on), gender, and type of job (paid or volunteer).

Regarding the job satisfaction, I have summed up their answers and labelled them as being satisfied, ambivalent and dissatisfied. Within the job satisfaction, I also measured their satisfaction on different aspects of their job (eg, pay, promotion) and each of these aspects summed up makes the job satisfaction score.

I am so confused on how to analyze the data. I have analyzed them using an independent samples t-test but I have just remembered how my supervisor told me I could use factor analysis. But I have no experience in using factor analysis. Can anyone please help me and advice me on how I should analyze my data?

@matt parker: thanks so much for your suggestion. I've only realized that I've done the wrong analysis, 1 week before the deadline! I'll try reading up on factor analysis now and fingers crossed that I'll get everything done by next week! :S

Answer 1

You have two questions:

• How to form an overall measure of job satisfaction?
• How to examine group differences on the score that you create?

Forming the overall measure of job satisfaction

If you are using an established measure of job satisfaction, then the test manual should tell you how you should calculate the overall job satisfaction score.

If the measure of job satisfaction is novel, there are multiple ways of forming an overall job satisfaction score where the individual items ask participants about facets of job satisfaction. In my experience, when you perform a factor analysis on a job satisfaction measure, the first unrotated factor explains a massive proportion of variance relative to any subsequent factors. As such, whether you run a factor analysis and save the first factor or whether you just take the mean of the items all measuring facet satisfaction, you are likely to be left with a very similar measure of overall job satisfaction (I'd expect correlations between the two forms to be in the r > .95 range). Of course you could and should test this idea in your data.

More importantly, there are general issues of validity. If you don't care too much about precision in measurement, then I would think that the first factor saved score or a mean of job satisfaction items would be a reasonable approximation to a measure of overall job satisfaction.

However, if you care about precision, you would want to engage with debates in the literature about whether overall job satisfaction should be asked directly rather than extracted from facet level measures. I discuss this a little more here.

Job satisfaction by group

Once you have your overall measure of job satisfaction, the task of comparing groups might look like this:

• For type of organisation, job status, and gender, independent groups t tests would work
• For age group, you could do an ANOVA with polynomial contrasts. In particular, if their is an effect of age it often has both linear and quadratic components. It would be better if you had a more granular measure of age.

Update

I received the following comment on my blog, where you wrote:

However, I am still confused of how my supervisor told me to use factor analysis but you seem to say that using the t test is enough. Can you please advise me further?

I am saying that you have two questions. The factor analysis pertains only to the first question of how to construct the overall measure of job satisfaction. After you have created that overall measure, whether it is informed by factor analysis or not, the tests of group differences are straightforward.

Answer 2

Your ultimate job satisfaction construct sounds suspect to me. When you summed up the different facets of job satisfaction to create your ultimate score, you implicitly assumed that each facet has the same weight - they really don't, as you'll see in the literature.

You'd have a much easier time with an explicit measure of overall satisfaction, which would let you model how much each facet contributes to overall satisfaction (plenty of examples of this in the literature). Without that, I think the best you could do would be to model individual facets of job satisfaction separately. I think that's where your supervisor's factor analysis suggestion comes in: if you've measured several facets of job satisfaction, some of them probably hang together. Factor analysis might let you create scales from those variables (e.g., "material rewards", "working environment"). I'm not the right person to give you advice on that, but I think that's probably your supervisor's general idea.

You sound like a university student, so find out if there's statistical support available for students and milk it for all it's worth. This stuff is not easy. That doesn't stop social scientists from doing it anyway, but just because your software will compute results and give you significant p-values for your model doesn't mean that model is worth a damn.

Answer 3

I am just learning statistical analysis and I am unsure if my response would be helpful but I just want you to stop for a while and think about what you are trying to achieve through your analysis. (Often in a state of panic, one usually ends up doing all sorts of analysis)

One of the most valuable advice I have received is that somethines you don't need complex statistical analysis to prove a point. It is fine if you are statstically trained but at the end of the day you must be able to understand and more importantly explain your methods and findings to others.

I would take a simple approach as follows:

1. Give a weight to the responses based on their relative importance/unimportance e.g. 1 point for statisfied, 0 for ambivalent and -1 for dissatisfied. (A likert scale response is better)
2. Ensure that your rating in (1) is equivalent i.e. have all three point or five point rating. This ensures the purity of the overall scale of the final scores.
3. Add the scores together

You now have an overall score for job satisfaction. You can run correlation analysis and then compare the coefficients to see if there is a difference.