Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Interpreting t-ratios for interaction effects in LCA

I've a question about the t-ratios in the Latent Class Analysis.  For the main effects holds if one of the levels of an attribute is significant (t ratio below  -1.6 or above 1.6), the attribute is significant and you can interpret the effects of the levels.
But how does this works for interaction effects?
In my case, duration (4 levels) and price (4 levels) makes 16 joint effects.

€30,- x 2 hours    0.45599    2.45755
€30,- x 4 hours    1.07400    -0.77317
€30,- x 6 hours    -1.01034    -0.46673
€30,- x 8 hours    -0.53699    -1.18404
€60,- x 2 hours    -0.37575    0.34347
€60,- x 4 hours    1.16512    1.33974
€60,- x 6 hours    1.09596    -0.42840
€60,- x 8 hours    -1.87651    -1.13654


Should one of the combinations of each fixed price level be significant or does something else apply?

Thanks in advance!
asked Dec 29, 2017 by anonymous

1 Answer

0 votes
I usually test the interaction, all 9 degrees of freedom, as a block:  run the model without them and then with them and see if your log likelihood improves significantly using the -2 log likelihood test.  If it does, keep them all in and if it does not, leave them all out.  

I usually don't pay much attention to the separate t-tests for each individual combination.
answered Dec 29, 2017 by Keith Chrzan Platinum Sawtooth Software, Inc. (93,025 points)