2

I'm seeking direction for the fit of a 5-point Likert scale attitudinal survey response to a 4-point scale that omits the center or Neutral response.

I am challenged by a very large corporate customer to assign and justify the assignment of the mid-range respondent answers to a 4-point model ranging from very satisfied to very dissatisfied. My results overall do not fit a normal distribution; rather, they skew strongly positive with the largest response frequencies being Very Satisfied (47% of total) and Satisfied (40% of total). The customer assumes that Neutral responses should be assigned to the negative, while I believe there is likely a proven approach to dividing the Neutrals between positive and negative satisfaction that mirrors the proportional split between the Very Satisfied and Satisfied versus the Dissatisfied and Very Dissatisfied.

Is there an approach that is supported by statistics and attitudinal research I should be following?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
K Kell
  • 21
  • 1
  • 5
    Could you explain why you need to perform this Procrustean feat of shoehorning a five-point response into a four-point scale? – whuber Mar 12 '18 at 20:42
  • 2
    The customer is a heavyweight in the aircraft and aerospace industry in the US. The expectation is to do what the customer asks you to do. I think the customer is better at building passenger jets than evaluating employee satisfaction with benefits. – K Kell Mar 14 '18 at 00:34
  • 4
    Regardless of the size of the customer, if they ask you to do something that cannot be justified or is outright wrong, the right response is not just to do it anyway! We all have (or have had) customers and bosses like that. As good consultants and employees we should seek to understand their needs and find solutions that are technically and ethically correct. Sometimes that means offering a creative alternative, perhaps by reinterpreting the request or finding a better solution. In this case you might begin by finding out the purpose behind the collapsed scale and its intended interpretation. – whuber Mar 14 '18 at 13:40

2 Answers2

1

Attitudes measures are infamous for displaying that negative skew that you are noticing. Generally this is because people:

A) Feel like the socially desirable thing to do is say you're happy.
B) Are generally happy, because if they weren't they would have left the organization and wouldn't be in your dataset

Generally it is not advisable to divide a continuum (or even a pseudo-continuum) into a dichotomous variable of satisfied or not satisfied. Its removing potentially important variance, and if you want to make any predictive model off of it, you will be shooting yourself in the foot. Because of this I dont think attitudinal research will have an answer to your problem, most researchers wouldn't see the point of removing variance.

If your client is pressuring you to do this though, you can divide it into "satisfied" (4 and 5) and "not satisfied" (1-3). Technically a neutral stance does mean that they are not "not satisfied" it also means that they are "not dissatisfied" though.

Hope that helps, good luck appeasing the customer!

NuclAcc
  • 66
  • 5
  • That makes sense and it would be my conclusion also. However, assigning Neutrals to a negative response count leaves me short of a 90 percent satisfaction mark. Missing the mark means a substantial financial penalty to the insurer. Customer and carrier have incentive to arrive at different performance outcomes. – K Kell Mar 14 '18 at 00:40
  • I guess if you really wanted you could calculate the proportion on the satisfied side, and the proportion on the dissatisfied, and divide the neutral by the proportion, but really it sounds like the insurer didn't hit their mark – NuclAcc Mar 14 '18 at 03:15
  • I was just thinking about this and its not quite what you asked but might solve your problem. You could significance test the proportion using a binomial test to see if its significantly different from .9. Alternatively calculate the standard error of the measure (SEM) and see if .9 is within the confidence interval – NuclAcc Mar 14 '18 at 14:20
  • Distributing the Neutrals between Satisfied and Dissatisfied according to the proportions of positive and negative responses makes practical sense but I wan't sure whether it might violate an established methodology. – K Kell Mar 14 '18 at 23:27
  • I think if you want to stick with a purely scientific approach you should go with the SEM to say its not significantly different from 90%. Theres not gonna be any scientific methodology on dividing the scale like this. – NuclAcc Mar 15 '18 at 03:41
1

This question seems to be about both modeling, ethics and cooperation/communication difficulties. There are some important comments that I cite here:

Could you explain why you need to perform this Procrustean feat of shoehorning a five-point response into a four-point scale?

– whuber

Regardless of the size of the customer, if they ask you to do something that cannot be justified or is outright wrong, the right response is not just to do it anyway! We all have (or have had) customers and bosses like that. As good consultants and employees we should seek to understand their needs and find solutions that are technically and ethically correct. Sometimes that means offering a creative alternative, perhaps by reinterpreting the request or finding a better solution. In this case you might begin by finding out the purpose behind the collapsed scale and its intended interpretation.

Apart from this, maybe you could try to avoid the problem with an analysis that in principle do not depend to much on how many Likert levels, such as ordinal regression. You could also present different analysis and how conclusions differ, with some simulations ...

The comments after the answer by@NuclAcc indicates financial interests in analysis results differ between interested parties. In that case, maybe even a court case is a possibility, and you want an analysis that you can defend in court.

– whuber

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467