

COMPLEX TWOLEVEL EFA processing time 

Message/Author 


I am attempting to run a complex two level EFA with 75 within items and 75 between items for 16 within factors and 16 between factors (N=2305; K=1; 14,203 missing item values). So far, my machine with a 64bit operating system and 5GB go RAM has not produced an output with my attempts of up to 6 hours in length. Is it feasible for me to run this analysis on a consumergrade computer? As a reference point, an identical analysis from the same data set on a smaller set of different items (25 within items and 25 between items for 6 within and 6 between factors; N=2305; K=1; 2,518 missing item values) takes approximately 10 minutes to produce an output on my machine. This is the analysis portion of my input: ANALYSIS: TYPE = COMPLEX TWOLEVEL EFA 1 16 1 16; ESTIMATOR = MLR; Thank you. 


Sorry, I did not accurately report how the data is clustered in my post. There are 207 clusters at the first level and 43 clusters at the second level. The full corrected post should say: I am attempting to run a complex two level EFA with 75 within items and 75 between items for 16 within factors and 16 between factors (N=2305; 207 clusters at the first level; 43 clusters at the second level; 14,203 missing item values). So far, my machine with a 64bit operating system and 5GB go RAM has not produced an output with my attempts of up to 6 hours in length. Is it feasible for me to run this analysis on a consumergrade computer? As a reference point, a similar analysis from the same data set with fewer items and fewer factors (25 within items and 25 between items for 6 within and 6 between factors; N=2305; 207 clusters at the first level; 43 clusters at the second level; 2,518 missing item values) takes approximately 10 minutes to produce an output on my machine. This is the analysis portion of my input: ANALYSIS: TYPE = COMPLEX TWOLEVEL EFA 1 16 1 16; ESTIMATOR = MLR; Thank you. 


I assume that your items are continuous. Sounds like your betweenlevel factors are defined by items that vary only on the between level, which would explain the unusually many betweenlevel factors. I wonder if the other items have variation also on the between level and therefore contribute to the measurement of the betweenlevel factors. To explore the slowdown, perhaps you could first try ... 1 16 1 1 to make sure the slowdown is not due to the within level. Then increase the betweenlevel factors. 


I tried 1 16 1 1 and it produced an output ‘INPUT READING TERMINATED NORMALLY’ with no numerical output. It took about 4 hours to produce that output. I also attempted to restrict the amount of output I request by performing a 12 12 12 12 analysis, resulting in the following output: THE H1 MODEL ESTIMATION DID NOT CONVERGE. SAMPLE STATISTICS COULD NOT BE COMPUTED. INCREASE THE NUMBER OF H1ITERATIONS OR THE H1CONVERGENCE OPTION. Increasing the number of H1ITERATIONS to 100000 produced the same message. Perhaps it will be helpful for me to explain the analysis in more detail. This is a diary study with students in science classrooms and all items we are currently dealing with in this analysis are measured at Level 1. But, we want to be able to justify using the measures aggregated up to Level 2 as well or know if we need to create different measures depending on the level of analysis. Hence, we are testing both daylevel (within) and personlevel (between) EFAs (adjusted using the complex option to account for the fact that people are also nested in 43 teachers) on the same 75 items. Person level variables are created by aggregating each person’s scores on the item across days and groupmean centering using teacher as the group. Day variables are groupmean centered using the person as the group. 


Perhaps you want to send your output and data to Support@statmodel.com so we can take a closer look at what's going on. 

Back to top 

