Hi Mike,
Hope this finds you well. I've been running moderator analyses using osmasem, and noticed that the parameter estimates from the model summary are not matching the parameter estimates you get when you request (LB) confidence intervals. This seems to only happen with models that include a moderator. The moderators are always centred. I feel like this is probably something simple, but I'm not sure. If I query the contents of A0, the parameters match the model summary, not the estimate from the confidence intervals function. Also, for some reason if I request intervals.type = "z", no confidence intervals are produced.
I've pasted the output for the intervals.type = "LB" model below.
Thanks!
Pasco
Summary of test with Parenting as moderator
free parameters:
name matrix row col Estimate Std.Error A z value Pr(>|z|)
1 parONrisk A0 2 1 -0.28413867 0.03052351 -9.308845 0.000000e+00
2 outONrisk A0 3 1 -0.27289355 0.02453159 -11.124168 0.000000e+00
3 outONpar A0 3 2 0.16974241 0.02139958 7.932043 2.220446e-15
4 parONrisk_1 A1 2 1 0.04386084 0.02799685 1.566635 1.172001e-01
5 outONrisk_1 A1 3 1 -0.03173868 0.02648485 -1.198371 2.307725e-01
6 outONpar_1 A1 3 2 -0.09168995 0.02245910 -4.082530 4.454801e-05
7 Tau1_1 vecTau1 1 1 -2.64736597 0.42141392 -6.282104 3.340219e-10
8 Tau1_2 vecTau1 2 1 -3.08032235 0.40981542 -7.516365 5.639933e-14
9 Tau1_3 vecTau1 3 1 -3.10602044 0.38516719 -8.064084 6.661338e-16
confidence intervals:
lbound estimate ubound note
test with Parenting as moderator.Amatrix[1,1] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Amatrix[2,1] -0.3366215156 -0.236482383 -0.149176322
test with Parenting as moderator.Amatrix[3,1] -0.3939021501 -0.307378694 -0.209565794
test with Parenting as moderator.Amatrix[1,2] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Amatrix[2,2] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Amatrix[3,2] 0.0020699564 0.070118191 0.137725321
test with Parenting as moderator.Amatrix[1,3] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Amatrix[2,3] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Amatrix[3,3] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Smatrix[1,1] 1.0000000000 1.000000000 1.000000000 !!!
test with Parenting as moderator.Smatrix[2,1] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Smatrix[3,1] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Smatrix[1,2] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Smatrix[2,2] 0.8866793258 0.944076083 0.977747512
test with Parenting as moderator.Smatrix[3,2] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Smatrix[1,3] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Smatrix[2,3] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Smatrix[3,3] 0.8327605324 0.890408045 0.939714056
test with Parenting as moderator.Tau2[1,1] 0.0005847094 0.005017959 0.020457527
test with Parenting as moderator.Tau2[2,1] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Tau2[3,1] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Tau2[1,2] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Tau2[2,2] 0.0002767570 0.002110892 0.010066294
test with Parenting as moderator.Tau2[3,2] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Tau2[1,3] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Tau2[2,3] 0.0000000000 0.000000000 0.000000000 !!!
test with Parenting as moderator.Tau2[3,3] 0.0002895904 0.002005141 0.008505026
test with Parenting as moderator.ind[1,1] -0.0636302136 -0.048230382 -0.034678023
Model Statistics:
| Parameters | Degrees of Freedom | Fit (-2lnL units)
Model: 9 38 -84.34805
Saturated: 9 38 NA
Independence: 6 41 NA
Number of observations/statistics: 6433/47
Information Criteria:
| df Penalty | Parameters Penalty | Sample-Size Adjusted
AIC: -160.3481 -66.348051 -66.32003
BIC: -417.5775 -5.425285 -34.02497
CFI: NA
TLI: 1 (also known as NNFI)
RMSEA: 0 [95% CI (NA, NA)]
Prob(RMSEA <= 0.05): NA
To get additional fit indices, see help(mxRefModels)
timestamp: 2020-01-18 17:24:34
Wall clock time: 2.592745 secs
optimizer: SLSQP
OpenMx version number: 2.15.4
Need help? See help(mxSummary)
Printer-friendly version
Hi Pasco,
Could you please post the data and R code so that I can take a look at it? Thanks.
Best,
Mike
Hi Mike,
Brilliant, thank you! Attached is the excel file and r script.
Best,
Pasco
Hi Pasco,
Thanks for posting the data and R code.
The osmasem() is wrapper of the original OpenMx functions. Unlike the tssem2(), intervals.type = "z" does not create the Wald CIs (yet).
When there are moderators, the intercepts and slopes parameters are stored in the A0 and A1 matrices. The Amatrix is the sum of these A0 and A1. The attached R code works for me.
Best,
Mike
Hi Mike,
Ahh, got it, thanks so much!
Best,
Pasco