| Message/Author |
|
|
|
|
Can LTA be performed only when the interval periods between two data points are the same in individuals used for data analyses?
For example, if some of the individuals are followed up 2 years after the first data collection and others are followed up 4 years after the first data collection, can we still conduct LTA using all of those individuals? |
|
|
|
|
| It depends on the subject-matter e.g. how quickly things change. But it would be safest to do the LTA separately for the 2 groups of individuals. |
|
|
|
|
Thank you, Dr. Muthen.
To clarify, subjects are individuals with developmental disorders, and variables are their emotional and behavioral challenges using scales (same scales twice - at baseline and at the first follow-up). Caregivers of these individuals returned follow-up questionnaires at different times, leading to several different interval periods (2 years, 3 years, 3.5 years, 4 years, for example). |
|
|
|
|
| So maybe there are 2 processes, one for subjects and one for caregivers. Perhaps it is only the caregiver data that is collected at unequal time intervals. |
|
|
|
|
| All data (both at baseline and at the first follow-up) were obtained only from caregivers. |
|
|
|
|
I see. Maybe you can consider for instance 5 time points
0, 2, 3, 3.5, 4
and use a missing data flag for the time points that a caregiver doesn't provide data. Assuming you have some people covering adjacent time points at all points, you should still be able to estimate the model. |
|
|
|
|
sorry, my explanation was not clear and made you misunderstand.
There are only two data points (baseline and the follow-up). Only caregivers provided their children's emotional and behaivoral problems using the same scale both times.
As the researchers allowed caregivers to provide the follow-up data anytime within 2- 4 year-periods following the baseline data collection, each child's data can have different intervals between those two time points. |
|
|
|
|
| Ok, so everyone has only 2 time points. There is no way to explicitly include the quantitative time difference in the modeling but you can capture it approximately by a multiple-group approach using Knownclass where the groups are created as subjects with approx. the same time interval. C2 ON C1 then naturally varies across the groups but the measurement part of the model can be group-invariant. |
|
|
|
|
| Thank you. My understanding is that the process (multi-group approach using Knownclass) is going to be done after determining the classes in LCA for each timepoint, right? My apologies - I am still learning basics of LCA and LTA. Thank you for your help. |
|
|
|
|
| Yes, it is useful to start with LCAs to support a decision on the number of factors. But you also need the LTA to try different number of classes - the LTA may suggest a different number of classes than the repeated LCAs. |
|
|
|
|
| Thank you so much for your advice. I will try that way. |
|
|
| Back to top |
