0

I am creating an Overall Customer Satisfaction Index score based off of 4 factors that comprise satisfaction for callers to a call center: A representatives concern for your needs; ease of navigating the phone system; usefulness of information provided; and promptness in speaking to a person. Scores for these subscales are rated on a likert scale (likert 1-10), and are then added together to comprise the Overall Satisfaction Index (composite score).

However, during the factor analysis, we learned that some factors explain variance more than others. So simply adding all the subscales (factors) together would be wrong, as some are more important to overall satisfaction (as I understand it). Therefor, a score of 10 on one subscale may be weighted higher on the index than a 10 on a subscale that explained less variance.

My question is how do I weight variables that differ in explained variance? Based on information online, I believe I used saved F-scores from these factors? However, I haven't a clue as to how I do this. Ultimately, I need to come up with a general multiplier weight based on the factor analysis that I can use with future call centers on the same index?

From comments:
I have 4 factors that determine customer satisfaction: Promptness; Usefulness; Ease of navigating; Concern for your needs. Each of these factors comprises 4 questions (selected using factor loadings). The questions are rating scales from 1-10. The mean is extracted for each subscale: Example Promptness - 7; Usefulness - 6; Ease of navigating - 5; Concern for your needs -10. The overall satisfaction score is the total added up: 28. However, we know that promptness explains more variance, so shouldnt this have a higher weight? Would I use the f-score as that weight?

This is an Overall Satisfaction score for only 1 call center. We will then run the same survey for a different call center. Ultimately we will have 100 call centers, each with different overall satisfaction scores in order to rank them.

gung - Reinstate Monica
  • 132,789
  • 81
  • 357
  • 650
Negative Correlation
  • 145
  • 8
  • 2
    This question seems to be unclear at present because of its inarticulate construction. Please describe in details what you did, what your aims are, what concerns you. – ttnphns Jun 27 '15 at 20:57
  • I hope this makes more sense – Negative Correlation Jun 28 '15 at 19:03
  • Asking 'how do I do ____ in SPSS?' is off topic here. If you want to convert this into a question about how to deal w/ factors that differ in explained variance, it may be reopened, & you may even (or may not) get help w/ the software usage elements. – gung - Reinstate Monica Jun 28 '15 at 19:06
  • The scenario still doesn't make sense, though. You *don't* add different factors together. They are different things. You wouldn't add someone's weight in kg + their height in cm to get a single (meaningless) score. In addition, factor analysis is typically followed by a rotation to enhance interpretability & which has the effect of largely equalizing the variances. Did you do that? You do add the scores from different items that load onto the same factor to get your measurement of that factor. Is the latter what you meant? – gung - Reinstate Monica Jun 28 '15 at 19:11
  • I've edited the question further – Negative Correlation Jun 28 '15 at 19:11
  • Yes, I did both varimax and promax rotations, both yielded similar results. When looking at overall satisfaction for call centers, this encompasses different things. The score is only meaningless in relation to itself, however it can be used to rank order other call centers on the same scale. This is the aim. – Negative Correlation Jun 28 '15 at 19:18
  • So is your question then about how to create a [multiple attribute ranking](http://stats.stackexchange.com/q/9358/7290), then? – gung - Reinstate Monica Jun 28 '15 at 19:23
  • I'm not sure that's the solution. In that case, the factors I extracted would be called attributes and then how would I combine them? – Negative Correlation Jun 28 '15 at 19:31
  • I'm just trying to understand your question. As stated, it doesn't make any sense to me (I'm not trying to be mean here). If you want to take several different factors & use them to rank different call centers, that sounds like you are asking how to create a multiple attribute ranking system. Are you? (You don't have to be, but if not...) Can you clarify what you are asking? – gung - Reinstate Monica Jun 28 '15 at 19:56
  • Sorry if im causing confusion. So i have 4 factors that determine customer satisfaction:Promptness;Usefulness;Ease of navigating; Concern for your needs. Each of these factors comprises 4 questions (selected using factor loadings). Questions are rating scales from 1-10. The mean is extracted for each subscale: Example Promptness - 7; Usefulness - 6; Ease of navigating - 5; Concern for your needs -10; The Overall Satisfaction Score is the total added up: 28 However, we know that promptness explains more variance, so shouldnt this have a higher weight? Would I use the fscore as that weight? – Negative Correlation Jun 28 '15 at 20:39
  • This is an Overall Satisfaction score for only 1 call center. We will then run the same survey for a different call center. Ultimately we will have 100 call centers, each with different overall satisfaction scores in order to rank them. – Negative Correlation Jun 28 '15 at 20:43

0 Answers0