Sat, 11/27/2010 - 09:40

Hello all,

when I want to fit my model, I receive this warning message:

In model 'Similarity Model' NPSOL returned a non-zero status code 6. The model does not satisfy the first-order optimality conditions to the required accuracy, and no improved point for the merit function could be found during the final linesearch (Mx status RED)

and when I then summary my fit, some values are missing.

Below, you can see, my model and also results from summary.

I am a begginer, so I would be really happy, if someone helped me.

Alena

simModel<-mxModel("Similarity Model",

type="RAM",

mxData(

observed=sim,

type="raw"

),

manifestVars=c("plot","depth","veg_div","sim","man","aver_L",

"aver_H",

"VVS","cover","GWT","pH","Ca","K","Mg","P","N","C"),

latentVars=c("vegetation","productivity","soil_depth","soil","SB",

"topography","disturbance"),

# residual variances

mxPath(

from=c("plot","depth","veg_div","sim","man","aver_L","aver_H",

"VVS","cover","GWT","pH","Ca","K","Mg","P","N","C"),

arrows=2,

free=TRUE,

values=c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1),

labels=c("e1","e2","e3","e4","e5","e6","e7","e8","e9","e10","e11",

"e12", "e13","e14","e15","e16","e17")

),

# latent variances and covariance

mxPath(

from=c("vegetation","productivity","soil_depth","soil","SB",

"topography","disturbance"),

arrows=2,

all=TRUE,

free=TRUE,

values=c(1,.5,.5,.5,.5,.5,.5,.5,1,.5,.5,.5,.5,.5,.5,.5,1,.5,.5,.5,.5,.5,

.5,.5,1,.5,.5,.5,.5,.5,.5,.5,1,.5,.5,.5,.5,.5,.5,.5,1,.5,.5,.5,.5,.5,.5,

.5,1),

labels=c("var1","cov12","cov13","cov14","cov15","cov16","cov17",

"cov12","var2","cov23","cov24","cov25","cov26","cov27","cov13",

"cov23","var3","cov34","cov35","cov36","cov37","cov14","cov24",

"cov34","var4","cov45","cov46","cov47","cov15","cov25","cov35",

"cov45","var5","cov56","cov57","cov16","cov26","cov36","cov46",

"cov56","var6","cov67","cov17","cov27","cov37","cov47","cov57",

"cov67","var7")

),

# factor loadings for vegetation

mxPath(

from="vegetation",

to=c("veg_div"),

arrows=1,

free=c(FALSE),

values=c(1),

labels=c("a")

),

#factor loadings for productivity

mxPath(

from="productivity",

to=c("cover","VVS","aver_L","aver_H"),

arrows=1,

free=c(FALSE,TRUE,TRUE,TRUE),

values=c(1,1,1,1),

labels=c("b","c","d","e")

),

#factor loadings for soil_depth

mxPath(

from="soil_depth",

to=c("depth"),

arrows=1,

free=c(FALSE),

values=c(1),

labels=c("f")

),

#factor loadings for soil

mxPath(

from="soil",

to=c("N","C","pH","K","Mg","Ca","P"),

arrows=1,

free=c(FALSE,TRUE,TRUE,TRUE,TRUE,TRUE,TRUE),

values=c(1,1,1,1,1,1,1),

labels=c("g","h","i","j","k","l","m")

),

#factor loadings for SB

mxPath(

from="SB",

to=c("sim"),

arrows=1,

free=c(FALSE),

values=c(1),

labels=c("n")

),

#factor loadings for topography

mxPath(

from="topography",

to=c("plot","GWT"),

arrows=1,

free=c(FALSE,TRUE),

values=c(1,1),

labels=c("o","p")

),

#factor loadings for disturbance

mxPath(

from="disturbance",

to=c("man"),

arrows=1,

free=c(FALSE),

values=c(1),

labels=c("q")

),

#means

mxPath(

from="one",

to=c("plot","depth","veg_div","sim","man","aver_L","aver_H",

"VVS","cover","GWT","pH","Ca","K","Mg","P","N","C","vegetation",

"productivity","soil_depth","soil","SB","topography","disturbance"),

arrows=1,

free=c(TRUE,TRUE,TRUE,TRUE,TRUE,TRUE,TRUE,TRUE,TRUE,TRUE,TRUE,

TRUE,TRUE,TRUE,TRUE,TRUE,TRUE,FALSE,FALSE,FALSE,FALSE,FALSE,

FALSE),

values=c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0),

labels=c("mean1","mean2","mean3","mean4","mean5","mean6",

"mean7","mean8","mean9","mean10","mean11","mean12","mean13",

"mean14","mean15","mean16","mean17",NA,NA,NA,NA,NA,NA,NA)

)

)

simFit <- mxRun(simModel)

Running Similarity Model

Warning message:

In model 'Similarity Model' NPSOL returned a non-zero status code 6. The model does not satisfy the first-order optimality conditions to the required accuracy, and no improved point for the merit function could be found during the final linesearch (Mx status RED)

summary(simFit)

data:

$`Similarity Model.data`

plot depth veg_div sim

Min. :0.0000 Min. :0.0 Min. : 6.00 Min. :0.0000

1st Qu.:0.0000 1st Qu.:0.0 1st Qu.:25.00 1st Qu.:0.1050

Median :0.0000 Median :0.5 Median :32.00 Median :0.1490

Mean :0.4928 Mean :0.5 Mean :30.43 Mean :0.1649

3rd Qu.:1.0000 3rd Qu.:1.0 3rd Qu.:37.00 3rd Qu.:0.2220

Max. :1.0000 Max. :1.0 Max. :51.00 Max. :0.4650

man aver_L VVS GWT

Min. :0.0000 Min. : 0.700 Min. :0.0120 Min. :-10.0

1st Qu.:0.0000 1st Qu.: 3.500 1st Qu.:0.1432 1st Qu.: 70.0

Median :0.0000 Median : 5.100 Median :0.2040 Median :125.0

Mean :0.3671 Mean : 6.316 Mean :0.2117 Mean :154.9

3rd Qu.:0.0000 3rd Qu.: 7.675 3rd Qu.:0.2580 3rd Qu.:216.5

Max. :3.0000 Max. :24.600 Max. :0.4990 Max. :707.0

pH Ca K Mg

Min. :4.220 Min. : 3.63 Min. : 2.450 Min. : 2.150

1st Qu.:5.010 1st Qu.: 53.58 1st Qu.: 7.335 1st Qu.: 6.537

Median :5.260 Median : 83.06 Median :10.240 Median : 9.680

Mean :5.279 Mean :109.17 Mean :13.271 Mean :13.667

3rd Qu.:5.548 3rd Qu.:147.91 3rd Qu.:14.668 3rd Qu.:15.930

Max. :6.800 Max. :478.16 Max. :78.520 Max. :73.530

P N C

Min. : 0.290 Min. :0.0850 Min. : 5.85

1st Qu.: 1.823 1st Qu.:0.2100 1st Qu.:10.42

Median : 2.130 Median :0.2520 Median :11.99

Mean : 2.923 Mean :0.2849 Mean :13.37

3rd Qu.: 3.172 3rd Qu.:0.3262 3rd Qu.:13.70

Max. :16.610 Max. :1.4220 Max. :70.15

The model does not satisfy the first-order optimality conditions to the required accuracy, and no improved point for the merit function could be found during the final linesearch (Mx status RED)

free parameters:

name matrix row col Estimate Std.Error

1 c A aver_L productivity 2.625390e+02 NaN

2 i A pH soil 2.935776e-01 1.129321e-02

3 l A Ca soil 9.552065e+01 4.482451e+00

4 j A K soil 4.946630e+00 NaN

5 k A Mg soil 1.106808e+01 NaN

6 m A P soil 3.832614e-01 1.153177e-01

7 h A C soil 3.885944e+00 1.411022e-01

8 p A GWT topography 1.600260e+02 6.841943e+00

9 e1 S plot plot 4.312332e-01 4.224270e-02

10 e2 S depth depth 4.983519e-01 NaN

11 e3 S veg_div veg_div 2.940395e+01 6.641467e+01

12 e4 S sim sim 2.970145e+00 NaN

13 e5 S man man 7.689308e-01 1.311142e+00

14 e6 S aver_L aver_L -1.402127e+01 NaN

15 e7 S VVS VVS 8.703063e-03 5.895986e-04

16 e8 S GWT GWT -7.595319e+01 NaN

17 e9 S pH pH 1.004841e-01 7.089877e-03

18 e10 S Ca Ca -8.869452e+01 NaN

19 e11 S K K 6.770639e+01 3.469788e+00

20 e12 S Mg Mg 8.308394e+01 5.045812e+00

21 e13 S P P 4.089212e+00 3.127822e-01

22 e14 S N N 6.785176e-01 NaN

23 e15 S C C 3.380603e+01 1.944256e+00

24 var1 S vegetation vegetation 2.940395e+01 6.641467e+01

25 cov12 S vegetation productivity 1.477552e-03 7.685325e-03

26 var2 S productivity productivity 6.573134e-04 4.748615e-05

27 cov13 S vegetation soil_depth -1.965396e+00 4.667837e-01

28 cov23 S productivity soil_depth -1.425096e-02 1.132874e-03

29 var3 S soil_depth soil_depth 4.967564e-01 NaN

30 cov14 S vegetation soil -4.143258e+00 1.015247e+00

31 cov24 S productivity soil -2.879921e-02 8.988565e-04

32 cov34 S soil_depth soil 1.507398e+00 NaN

33 var4 S soil soil 3.748584e+00 2.324992e-01

34 cov15 S vegetation SB -3.906109e-01 9.030622e-02

35 cov25 S productivity SB -1.681545e-03 1.369041e-04

36 cov35 S soil_depth SB 9.102603e-02 9.106335e-03

37 cov45 S soil SB 1.985258e-01 2.762651e-02

38 var5 S SB SB -2.950044e+00 NaN

39 cov16 S vegetation topography 7.189850e+00 1.404624e+00

40 cov26 S productivity topography 4.651940e-02 3.142341e-03

41 cov36 S soil_depth topography -2.327834e+00 NaN

42 cov46 S soil topography -4.859159e+00 NaN

43 cov56 S SB topography -3.294746e-01 3.669061e-02

44 var6 S topography topography 8.538890e+00 NaN

45 cov17 S vegetation disturbance -4.054148e+00 5.953045e-01

46 cov27 S productivity disturbance -4.890707e-03 1.194785e-03

47 cov37 S soil_depth disturbance 3.857770e-01 1.264814e-01

48 cov47 S soil disturbance 1.040433e+00 2.137051e-01

49 cov57 S SB disturbance 5.993311e-02 1.731205e-02

50 cov67 S topography disturbance -1.810321e+00 1.286270e-01

51 var7 S disturbance disturbance 7.668210e-01 1.264187e+00

52 mean1 M 1 plot 7.544507e-02 1.522074e-01

53 mean2 M 1 depth 5.917900e-01 5.138060e-02

54 mean3 M 1 veg_div 3.009372e+01 3.858053e-01

55 mean4 M 1 sim 1.779674e-01 7.327289e-03

56 mean5 M 1 man 4.748542e+02 NaN

57 mean6 M 1 aver_L 5.669472e+00 2.876393e-01

58 mean7 M 1 VVS 2.066480e-01 4.769057e-03

59 mean8 M 1 GWT 8.357171e+01 2.459515e+01

60 mean9 M 1 pH 5.334432e+00 3.214698e-02

61 mean10 M 1 Ca 1.272341e+02 9.365325e+00

62 mean11 M 1 K 1.408931e+01 6.083485e-01

63 mean12 M 1 Mg 1.599236e+01 1.140197e+00

64 mean13 M 1 P 3.001567e+00 1.082431e-01

65 mean14 M 1 N 4.462727e-01 1.053452e-01

66 mean15 M 1 C 1.406673e+01 4.644328e-01

67

observed statistics: 6210

estimated parameters: 67

degrees of freedom: 6143

-2 log likelihood: 28740.37

saturated -2 log likelihood: NA

number of observations: 414

chi-square: NA

p: NA

AIC (Mx): 16454.37

BIC (Mx): -4138.261

adjusted BIC:

RMSEA: NA

timestamp: 2010-11-27 13:47:57

frontend time: 3.421 secs

backend time: 1.12475 mins

independent submodels time: 0 secs

wall clock time: 70.906 secs

cpu time: 70.906 secs

openmx version number: 1.0.3-1505

I agree with Ryne that identification problems are the likely cause of Code Red, IFAIL 6 on this occasion. However, note also that Code Red isn't always as disastrous as it sounds - sometimes the error can be reported even if optimization has identified the global minimum. So code reds can be false alarms. Here's some text from another thread about what to do when it is encountered.

********************************************

Optimization may not have been successful, NPSOL IFAIL= 6. For a full technical description of what this means, please see http://www.sbsi-sol-optimize.com/asp/sol_product_npsol.htm and read the notes below.

There are several things to note about IFAIL = 6

1. Often, but not always, IFAIL = 6 is returned when optimization has completed successfully.

2. It is still the responsibility of the user to check that the solution is a global minimum. Unfortunately, optimization is not an exact science.

3. Checking that the solution is a global minimum can be done several ways:

i) Fit the model again from the present solution several times. If the fit improves, and keeps improving over the successive optimizations then a global solution probably has not been found, and the number of refits should be increased until the solution stabilizes. Here is some example R code that can be used to refit the model:

refits <- 3

for(i in 1:refits) {

model <- mxOption(model, "Cold Start")

model <- mxRun(model)

summary(model)

}

ii) Try starting optimization from a different place by changing the starting values of the parameters. These could be manually changed, but note that changing all the starting values from say .6 to .7 may not be sufficient. Random starting values could be assigned using, e.g., rnorm(), but be careful to keep the range of possible starting values reasonable.

iii) Consider whether the model is identified. IFAIL = 6 does occur more often with underidentified, or nearly-identified models. If two different sets of parameter estimates yield the same -2lnL (or whichever fit function is being used) then the model can be said to be underidentified. Sometimes this condition arises because of characteristics of the data, particularly if the dataset does not provide much information to estimate every parameter of the model.

Pretty complicated model for a beginner. Way to go!

The "Code Red" error can happen for a variety of reasons. The first thing I do is look at the variance estimates to make sure they are positive. In your case, you have a number of negative variance estimates:

14 e6 S aver_L aver_L -1.402127e+01 NaN

16 e8 S GWT GWT -7.595319e+01 NaN

18 e10 S Ca Ca -8.869452e+01 NaN

38 var5 S SB SB -2.950044e+00 NaN

For the residual variances (parameters 14,16, & 18), I might constrain the lower bound of the parameters with an lbounds=.00001 argument in your mxPath statement that creates the residual variances. Doesn't hurt to make that constraint for all the residual terms at once, so just add "bounds=.00001" to that one mxPath statement.

This might be enough. You also have a fairly large negative variance estimated for one of your latent variables. But it is only indicated by one manifest variable and you have a fixed loading of 1 for it. So, I'd just add an lbound to that mxPath as well.

Hope that gets you going!

Thank you for your reply. Now, I have not so complicated models and I tried your advice and added lbounds to residual variances and then as well to latent variances. But I had still the same warning message as before.

Maybe, the problem can be in specifying fixed loading values and using free as TRUE or FALSE, because I am not sure with my decisions.

I think I see your problem, and its that your model isn't identified. You have a number of factors indicated by a single manifest variable. This is fine, but you cannot simultaneously estimate a residual variance and a factor variance without extra information from other variables. If you fix the residual variances at zero for any manifest variables that have exactly one relationship (i.e., they are the only manifest variable for a factor, and have no relationships but with that factor), then your model will be identified.

While you specify them as latent variables, there is no distinction between a manifest variable and latent variable that fully explains that manifest variable. You can only estimate one free parameter for that manifest-latent combination. Your model was unidentified (leading to the NPSOL 6 warning) because for every possible value for the variance of the latent variable, you can find a value for the residual variance that doesn't affect fit. Fix one of the parameters, and the other parameter becomes identified. Does that make sense?