latent class - initial values not feasible - tried several option

[Rose Matthews]

I've got a task -the ability to go about the stairs-
This was measured -BEFORE- and -AFTER- treatment

Therefore I have 2 variables: preop_stairs , postop_stairs - both are ordinal (1-5), 5 being the worse

I want to run a -LATENT CLASS ANALYSIS- to determine if there are patterns eg:
-Early improvers
-Late improvers

As you may have realised in a previous post I've had a couple of problems, initially I tried to combine the -preop_* and postop_* for 4 different tasks (that would mean 8 variables) but it's caused too many problems being so slow. I've kept the analysis running for a total of 15 hours and it wasn't reading.

I've decided to separate the tasks out, therefore only dealing with 2 variables (as seen above). I thought this would solve the problems, but still causing problems
I'm not at a point trying to decide whether to proceed with latent class analysis as it's causing too many problems :


Dataset originally: 265,000 observations

I ran:

gsem (preop_stairs postop_stairs <- ologit), lclass(C 3)

////Error: initial values not feasible

gsem(preop_stairs <-, ologit) (C <- postop_stairs), lclass(C 3) iterate(12)

////Error: initial values not feasible

I then tried:

[CODE
gsem(preop_stairs <-, ologit) (C <- postop_stairs), lclass(C 3)
[/CODE]

////Error: initial values not feasible

I then tried , taking a random sample, and ran the following

sample 10, by(gender)
gsem (preop_stairs postop_stairs <- ologit), lclass(C 3)
gsem(preop_stairs <-, ologit) (C <- postop_stairs), lclass(C 3) iterate(12)
gsem(preop_stairs <-, ologit) (C <- postop_stairs), lclass(C 3)

////Error: initial values not feasible


What is your advice?