Single-item versus multi-item measures of job satisfaction
A measure of job satisfaction is good if it is valid. By valid, I simply mean that there is a true latent variable called job satisfaction and that the measured variable correlates highly and has a small squared measurement error with the true latent variable.
In order for a measure to be valid, it needs to be reliable.
There is a debate in the literature about whether single item measures of job satisfaction are adequate.
Landy and Conte (p. 412) write:
"Wanous, Reichers, and Hundy (1997) demonstrated that even single items of job satisfaction (e.g., "Overall, how satisfied are you with your current job?") may work well in many situations."My thoughts:
Validity first requires reliability. To define reliability we need to speak about the level of analysis. Reliability at group-levels of analysis, such as the organisation or department, is very different to reliability at the individual-level. Reliability will almost always be greater at the organisational level than it will be at the individual level. The more reliable a measure is at the individual-level, the more reliable it will be at the group-level. However, group-level measurement becomes more reliable as the number of individuals in the group increases.
I have not yet got round to articulating the specific mathematical formulas which would govern such a relationship, but my expectation would be that given sufficient sample (e.g., n > 1000), even somewhat unreliable measures at the individual level (e.g., alpha = .5) would be highly reliable at the group-level.
Additional issues might arise related to non-response. Thus, in estimating organisation-level job satisfaction, there is the uncertainty introduced as a result of participants who do not respond to the survey.
Reliability at the individual-level will generally be higher if the variable is based on multiple items. Two reasons for this are, first, that a multi-item scale is likely to be more nuanced and will have more possible scale points, and second that observing multiple behaviours (i.e., multiple responses to survey items) will reduce the effect of randomness in responding. And to reiterate, improved reliability at the individual level will improve reliability at the group-level, but at a certain sample size, such improvements are likely to be of minimal influence.
Implications:
The implications of the above discussion are that single items will typically be very reliable measures at the organisational-level (at least where sample size is fairly large). Thus, reports of organisational-level mean scores for an individual item or the percentage of employees that endorse a particular item will tend to be highly reliable. And thus, it is reasonable to use single items to do analyses at the organisational-level such as comparing mean scores from one year to the next or comparing mean scores between a benchmark and the focal organisation.
However, if the interest is on the individual-level of analysis, using individual items will tend to be more problematic. There is some research on the reliability of single items (e.g., Dolbier et al, see here). While the reliability of single items might be reasonable, it is likely (i.e., I still need to check what the empirical literature says on this matter) that well constructed multi-item scales will be of greater reliability. Thus, if you want to run analyses that correlate job satisfaction with other variables at the individual-level, a job satisfaction scale is the better option. Likewise, if you want to model individual job satisfaction change trajectories, job satisfaction scales are better.
Validity of an overall measure versus one computed from facets:
Landy and Conte (p. 412) write:
"... many researchers [take] the position that overall satisfaction is the result of combining satisfaction with specific important aspects of work. Thus, they would advocate using a mathematical formula for weighting and combining satisfaction with specific aspects."Landy and Conte then go on to present Judge and colleagues (2001) five-item scale measuring overall job satisfaction, and the Job Descriptive Index as a popular measure that incorporates facets.
My thoughts:
Conceptually, overall job satisfaction represents an overall appraisal of a job by an individual. Rationally, such an appraisal should be influenced by satisfaction with facets of a job, with greater weight given to facets of greater importance. But what if people do not appraise the job in such a way? What if, hypothetically, satisfaction with facets has nothing to do with overall job satisfaction? Would we say that people are incorrect in their overall appraisal, or would we say that the construct of overall satisfaction is separate from any mathematical function of facet satisfaction? And if they were distinct, which variable would we prefer, the mathematical function of facets or the overall measure?
Of course, most of this discussion is academic, because from my experience, overall job satisfaction measures tend to be have fairly high correlations with mathematical functions of facet-based measures. However, I am inclined to prefer a measure such as the five-item scale developed by Judge and colleagues (2001), if my intention was to measure overall job satisfaction. If people base their overall satisfaction on importance-weighted-satisfaction with facets then the overall satisfaction measure should capture this implicitly. The overall measure lets the employee decide how this function should apply and removes the need to try to estimate what this function is for a particular individual.
However, It is still useful to measure satisfaction with facets, such as pay, supervision, coworkers, and so on. This would not be as a means of measuring overall satisfaction. Rather it would be because monitoring facet satisfaction provides useful information to organisations to monitor their workforce and guide action.
Equally measuring facet satisfaction in various ways might be useful in research that aims to understand the information processing that governs the appraisal process on the formation of overall job satisfaction (see for example, Lawler's 1973 model - p. 408 Landy and Conte).
I hope to continue this discussion in a future post focusing particularly on the issue of the role of job satisfaction in the causal system of the organisation.
