

Missing Categorical Predictors in Mul... 

Message/Author 


Hi there, I am doing a multilevel path model, which predicts how teachers rate a student through student sex, SES (4 different categories, split into 3 dichotomous variables), and IQ at Level 1, and class size and classaverage IQ at Level 2. I have missing data on almost all predictors, between 5 and 28%. 1) How can I estimate these missing values for the multilevel analysis? 2) Ideally, the classaverage IQ should be based on the individual IQ values we have as well as the estimated missing individual IQ values. Is there a possibility to have this information entered automatically into the data file? Thanks, Tanja 


1. You can mention the variances of the covariates in the MODEL command. For covariates on the WITHIN list do this in the within part of the model. For covariates on the BETWEEN list do this in the between part of the model. For covariates on neither list, do this in both parts of the model. 2. See the CLUSTER_MEAN option of the DEFINE command. 


Hi Linda, 2) is fabulous. Thanks a lot for this hint, which saves me so much time! I did not understand your first point, though. Sorry. Could you maybe give me a brief syntax example? This might make it clearer what you mean by "mention the variances". In how far would this help estimate the missing values? And an additional question on DEFINE: I have IQ raw values for children at grade level 1, 2, and 3 (rwiq) and would like to standardize this variable using the STANDARDIZE command, but for each grade level separately, such that it results in the same variable (e.g., "zrwiq") for the entire sample, as I want to know how a child achieves in relation to children of the same age only. Is this possible, and if so, how? Thanks for your patience, Tanja 


WITHIN=x1; BETWEEN=z; MODEL: %WITHIN% y ON x1 x2; x1 x2; ! variances of x1 and x2 %BETWEEN% y ON z x2; z x2; ! variances of z and x2 I'm not sure I understand the second part of your question. 


Hi Linda, thanks! This is very helpful! Somehow my first answer was not posted, I'll try again. The idea is that usually, you standardize, e.g., raw values of an IQ test into the IQ scale and do so for different age groups, as results that are perfectly average for a firstgrader are far below average for a third grader. The variable would mean something like "how well does the kid do, compared to age peers?" Usually, one would probably do the standardization for each subgroup and then merge the files. I was just wondering whether this can be done in one step using Mplus. Best, Tanja 


You can standardize for each age group using a conditional statement. See the DEFINE command. 

Back to top 

