Message/Author 


I am planning to do an EFA of parenting variables from a literacy study. The data is nested (within classroom and intervention type). I have a total of 90 variables that will be used in the EFA. Some are yes/no (coded 1 or 2), some are 17 Likert scale, some are 15 Likert scale, some are 13 Likert scale some are in minutes ranging from 1720. Is it possible to run an EFA with this many types of variables? I have reverse coded everything so that is all in the positive (higher means better). 


I think you are saying that students are nested within classrooms and that the intervention is given at the classroom level. If so, this is possible. I would rescale any continuous variables by dividing them by a constant to keep their variances between one and ten. 


yes that is correct, although there were 4 different interventions groups and 1 control. Parents received parenting education in all, but what and how they received it differed depending on the intervention. I am not looking at intervention effects, so does the fact that the data is nested even matter? 


How many classrooms do you have and what is the average class size? How many in each intervention group? With a good number of classrooms, it would be of interest to do a TYPE=BASIC TOWLEVEL analysis to see what your intraclass correlations are. The EFA may be influenced by high icc's and there are twolevel EFA alternatives. 


117 classrooms ranging in size from 1 to 35. 250 to 300 in each intervention. My N is 1400. Although, I may run an EFA on half and a CFA on the other half? 


Might be a good idea to as a first step do Type = Basic Twolevel and for simplicity treat all variables as continuous. You may want to do your EFA on separate sets of variables if they are designed to measure different constructs. 


Thank you so much for your help! The variables I am using come from 4 different sources. First is a parent interview (most of these are yes/no questions with some questions about how many hours/mins per day do you spend reading with your child, etc. I also have variables that were obtained via observations. Of those some are was this observed, yes or no. Others are on a 15 Likert scale (minimal to extensive). Others are number of times (17) a certain behavior was observed. All of these questions are related to parenting. Through the EFA, I want to be able to find out if there are any underlying factors among these 90 variables. Is this something that I can do in Mplus? 


Mplus can carry out such an analysis, but EFA is really meant for a set of items that were developed to measure certain dimensions. It is used to confirm that the items measure the dimensions as intended. See the Topic 1 course handout on the website where using EFA is described. 


is there something other than EFA that would be better to answer my question? what type of EFA would you recommend I run, given that I have nominal and interval scaled items within a nested dataset? 


You may want to ask this question on SEMNET. 


I am planning to run an EFA as I have mentioned followed by a CFA. I am planning to use the same data set, just half for the EFA and half for the CFA. Is it possible to run both from the same data file or do I need to have separate files for the data for EFA and the data for CFA? 


Another question to clarify...if I am using nominal dichotomous and interval data in my analysis, how to do I specify which variable is which in Mplus? 


You will need to use another program to randomly split your sample into two data sets. You would specify the dichotomous variables on the CATEGORICAL list. See the user's guide for further information. You do nothing special for the interval variables. 


Thank you! I have already split my sample in SPSS, but can I run both the EFA and CFA in Mplus from the same data file? 


If you want the data in one file, you will need a way to select the half you want. It seems easier to use different data sets if that is what you currently have. 


I have converted an SPSS file to .dat file. When I try to run Mplus, it says that one of the variables contains noninteger values. The data in SPSS is fine, but the .dat file has missing values in the last 4 variables. Am I doing something wrong when I convert the data? I am going to save as and selected tab delimited. I am unchecking write variable names to spreadsheet and I am choosing the variables to save. I also tried saving as a .csv file. When I did this it had an error message that said nonmissing blank found in data file at record #123, field #: 12. When I looked at the data, there is nothing missing? 


You cannot have blanks in the data set and use free format. SPSS seems to use blanks as a certain type of missing. You need to change that. If you can't see the problem, send the input, data, output, and your license number to support@statmodel.com. 


I am using a restricted use data set and cannot send the data. There are no blanks in the SPSS data just values (03). It still says that the categorical variable contains nonintegers? In addition a new error is coming up that says a categorical variable X has less than 2 categories. But it has values for 0 and 1? 


It is not the SPSS data set you should be looking at. You should look at the data set named in the FILE option. All messages point to you reading the data incorrectly either by having blanks in the data set or by the number of variable names in the NAMES statement not being the same as the number of columns in the data set. 

Back to top 