Anonymous posted on Wednesday, September 17, 2003 - 1:30 am
I conducted EFA based on Pearson correlation with 7-point scale items. However, journal reviewers asked me to conduct EFA treating my data as ordinal and to use a special program such as Mplus or MICROFACT. I understand their point but I assumed that 7-point would be enough for EFA using Pearson correlation. What do you think? Am I correct? If so, could you please introduce any references? Thanks so much in advance.
It depends on whether your variables have strong floor or ceiling effects. If they do, you most likely need to use a method that treats them as categorical not continuous variables. If they do not, you are most likely ok using Pearson corrleations. I don't know of any references specific to this topic. You could generate data like yours and analyze it using Pearson correlations and see how well you recover your parameter estimates. Or you could just analyze the data both ways and see what if any differences you find.
JeremyMiles posted on Wednesday, March 24, 2004 - 6:38 am
A paper related to this issue:
Hsu, T. and L. Feldt (1969). "The effect of limitations of the number of criterion score values on the significance level of the F-test." American Educational Research Journal 6: 515-527.
I prefer treating likert scale variables as ordered categorical versus continuous. When there are no strong floor or ceiling effects, the results may be close when treating them as continuous. Model estimation is computationally more demanding when treating them as categorical depending on the number of items in the analysis.
Theda Radtke posted on Wednesday, February 24, 2010 - 7:21 am
Dear Linda and Bengt Muthen, I`m also interested in the problem of the message MPlus gave me: NO CONVERGENCE. NUMBER OF ITERATIONS EXCEEDED. PROBLEM OCCURRED IN EXPLORATORY ANALYSIS WITH 4 FACTOR(S).
Because I had to analyze 5 data sets, it would be helpful if you can tell me possible reasons for that warning. Is the reason a distribution of the data or something like that? Thank you.
I am new to mplus and have a data set with 45 items on a 7 point scale (strongly disagree to strongly agree). My data set is 610 participants who make up a clinical (137) and non-clinical sample (473). My measure would be used to identify thoughts more readily endorsed by the clinical group. I would like to do EFA and have made the following choices, could you give me your opinion on each?
1. I have conducted EFA on the whole sample rather than just the clinical group for whom the items were chosen given the smaller sample size of the clinical group.
2.I have chosen to use WLSM and define my data as categorical because some (but not all of my items) show floor effects.
3.I have chosen the default rotation geomin the factors are likely to correlate.
4. I am choosing a factor structure in which RMSEA is .06 but CFI is .96 and SRMR is .035 in which the loadings make the most sense. What would be the reason to weight RMSEA higher in the decision process than CFI or SRMR?