|
|
How many indicators in LCA/LPA are to... |
|
Message/Author |
|
Anna V posted on Saturday, September 15, 2018 - 5:55 pm
|
|
|
Hi all, I will be conducting an LPA for someone, but their intended model has 50 continuous indicators, which seems like too many to me. Unfortunately, I can’t find any sources discussing such a large number of indicators. Would this be appropriate to do? Also, what would you consider an appropriate sample size for this? Thank you! |
|
|
50 is ok if N is large enough. The required N depends on how many classes you expect because this influences how many parameters you need to estimate. With k classes, you have 50 variances, k*50 means, and k-1 class probabilities. You might want at least 5 observations per parameter. |
|
Anna V posted on Monday, September 17, 2018 - 7:41 am
|
|
|
Thank you, this is very helpful! |
|
Ads posted on Wednesday, September 09, 2020 - 9:10 am
|
|
|
Is there a specific motivation for 5 observations per parameter? Also, is there an equation that shows the sample size requirement to literally identify an LPA model, given a specific number of indicators and classes? Why I ask: a colleague in neuroscience wants to estimate up to 4 classes with 75 indicators and N=150 via LPA. While that sounds like it is pushing the limits to me, I am literally able to simulate the model (and even able to simulate a model with 200 indicators and N=150, which would have many more parameters than participants). Clearly you could not have C>N, but I'm not sure how sample size would limit number of indicators used. In one Mplus forum post it's mentioned that there needs to be more subjects than parameters (http://www.statmodel.com/discussion/messages/13/489.html?1424463211), but my simulations seem to permit more parameters than subjects. Many thanks for your help. |
|
|
No, there is no such formula with continuous outcomes. It all depends on the distribution of the outcomes. If you have clear class separation in the outcome means, I can imagine that your case can work. |
|
Ads posted on Wednesday, September 09, 2020 - 5:24 pm
|
|
|
Sounds good - thanks again for your feedback. |
|
Back to top |
|
|