eaalogit to account for attribute non-attendance

[Eva Boonaert]

Hello

I am trying to account for attribute non-attendance for my choice experiment via the eaalogit command by Hole, Arne, (2016), EAALOGIT: Stata module to estimate endogenous attribute attendance models, https://EconPapers.repec.org/RePEc:boc:bocode:s457903.

I am using the following code:

matrix b0=e(b) #after I estimated my latent class model via lclogit
eaalogit chosen fpr fsec ftert mnopr mpr msec mtert fnopr_nch fpr_nch fsec_nch ftert_nch mnopr_nch mpr_nch msec_nch mtert_nch fnopr_nch2 fpr_nch2 fsec_nch2 ftert_nch2 mnopr_nch2 mpr_nch2 msec_nch2 mtert_nch2, group(choicecard) id(respondent) keaa(23) eaaspec(x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 x11 x12 x13 x14 x15 x16 x17 x18 x19 x20 x21 x22 x23) from(b0)

I tried this on three versions of Stata. However I get the following error messages on the two versions:
Version 14
initial vector: extra parameter choice1:fpr found
specify skip option if necessary
r(111);
Version 15
Some variables are collinear - if this is intended use the coll option. However I can't find how to use this command online.
Version 16
convergence not achieved
r(430);

Anyone who has an idea how to solve this?

Thank you very much in advance!

Kind regards

Eva

[Andrew Musau]

Version 14
initial vector: extra parameter choice1:fpr found
specify skip option if necessary
r(111);

This is usually solved with

from(b0, skip)

[Arne Risa Hole]

I find it odd that you are getting these different error messages in different versions of Stata. Are you sure that your code is identical and that you are using the same version of eaalogit in all three versions?

You don't show the code used to generate the starting values, but I suggest that you use the default starting values instead.

Consider simplifying the model. You are trying to estimate 23 ANA parameters which is asking a lot from your data. For example it is possible to assign the same ANA parameter to several variables (see the help file for an example), which is particularly useful if the variables are dummies representing different values of the same categorical attribute in the choice experiment.

[Eva Boonaert]

Andrew Musau and Arne Risa Hole, thank you for your answers!

I tried it for a reduced model as follows:

eaalogit chosen fpr fsec ftert mnopr mpr msec mtert, group(choicecard) id(respondent) keaa(8) eaaspec(x1 x2 x3 x4 x5 x6 x7 x8)  from(b0, skip)

However, then I get this error message:

               eaa_gf0():  3200  conformability error
      mopt__calluser_v():     -  function returned error
      opt__v0_calluser():     -  function returned error
    deriv__call1user_v():     -  function returned error
 _deriv__compute_value():     -  function returned error
                _deriv():     -  function returned error
       opt__eval_nr_v0():     -  function returned error
             opt__eval():     -  function returned error
opt__looputil_iter0_common():     -  function returned error
opt__looputil_iter0_nr():     -  function returned error
          opt__loop_nr():     -  function returned error
            _moptimize():     -  function returned error
           Mopt_maxmin():     -  function returned error
                 <istmt>:     -  function returned error
r(3200);

[Arne Risa Hole]

It is difficult to diagnose problems like this without seeing any of your data. Note that the group variable has to be a unique identifier for each choice occasion, so each choice occasion needs a different value (both within and across individuals).

As suggested try using the default starting values (drop "from(b0, skip)" from the syntax).

Your model specification is inconsistent - you have 7 explanatory variables and 8 ANA parameters.

A general recommendation based on reading this and some of your other recent posts is to start with a simple model and build up complexity gradually. Make sure you can get sensible results using e.g. asclogit first and then move on to more advanced models in incremental steps. Good luck.

[Eva Boonaert]

Arne Risa Hole thank you very much for your reply! My model above was indeed misspecified.

I performed a choice experiment with 12 cards per respondent each consisting of 2 alternatives. The model that I use (based on the AICc and BIC criteria and some assumptions to obtain a realistic model) includes 8 schooling components (fnopr, fpr, fsec, ftert, mnopr, mpr, msec and mtert) which are all interacted with the number of children (nch) and the number of children of power 2 (nch2). One of the main effects of the schooling components (fnopr) is omitted to allow to estimate my model since my model involves a mixture model.These are the first rows of my data:

 
respondent
choice card
time
alternative
chosen
fnopr
fpr
fsec
ftert
mnopr
mpr
msec
mtert
nch
nch2
nch3
nch4
nch5
nch6
nch7
nch8
fnopr_nch
fpr_nch
fsec_nch
ftert_nch
mnopr_nch
mpr_nch
msec_nch
mtert_nch
fnopr_nch2
fpr_nch2
fsec_nch2
ftert_nch2
mnopr_nch2
mpr_nch2
msec_nch2
mtert_nch2

1
1
1
1
1
0
0
0
0
0
0
1
0
1
1
1
1
1
1
1
1
0
0
0
0
0
0
1
0
0
0
0
0
0
0
1
0

1
1
1
2
0
0.333333
0.666667
0
0
0
0
0
0
3
9
27
81
243
729
2187
6561
1
2
0
0
0
0
0
0
3
6
0
0
0
0
0
0

1
2
2
1
0
0
0
1
0
0
0
0
0
1
1
1
1
1
1
1
1
0
0
1
0
0
0
0
0
0
0
1
0
0
0
0
0

1
2
2
2
1
0
0
0.25
0
0
0
0.25
0.5
4
16
64
256
1024
4096
16384
65536
0
0
1
0
0
0
1
2
0
0
4
0
0
0
4
8

1
3
3
1
0
0
0
0
0
0.333333
0
0.666667
0
3
9
27
81
243
729
2187
6561
0
0
0
0
1
0
2
0
0
0
0
0
3
0
6
0

1
3
3
2
1
0
0
0
0
0
0.666667
0.333333
0
3
9
27
81
243
729
2187
6561
0
0
0
0
0
2
1
0
0
0
0
0
0
6
3
0

1
4
4
1
0
0
0
0
0
0.333333
0
0
0.666667
3
9
27
81
243
729
2187
6561
0
0
0
0
1
0
0
2
0
0
0
0
3
0
0
6

1
4
4
2
1
0
0
0
0.6
0.4
0
0
0
10
100
1000
10000
100000
1000000
10000000
1E+08
0
0
0
6
4
0
0
0
0
0
0
60
40
0
0
0

1
5
5
1
0
0.333333
0.333333
0.333333
0
0
0
0
0
3
9
27
81
243
729
2187
6561
1
1
1
0
0
0
0
0
3
3
3
0
0
0
0
0

1
5
5
2
1
0.25
0
0.25
0
0
0
0
0.5
4
16
64
256
1024
4096
16384
65536
1
0
1
0
0
0
0
2
4
0
4
0
0
0
0
8

1
6
6
1
1
0
0
0
0
0
0
0.5
0.5
2
4
8
16
32
64
128
256
0
0
0
0
0
0
1
1
0
0
0
0
0
0
2
2

1
6
6
2
0
0
0
0
0.75
0.25
0
0
0
8
64
512
4096
32768
262144
2097152
16777216
0
0
0
6
2
0
0
0
0
0
0
48
16
0
0
0

1
7
7
1
1
0
1
0
0
0
0
0
0
3
9
27
81
243
729
2187
6561
0
3
0
0
0
0
0
0
0
9
0
0
0
0
0
0

1
7
7
2
0
0.2
0.2
0
0
0.2
0
0
0.4
10
100
1000
10000
100000
1000000
10000000
1E+08
2
2
0
0
2
0
0
4
20
20
0
0
20
0
0
40

1
8
8
1
0
0.2
0
0.4
0
0.2
0
0.2
0
5
25
125
625
3125
15625
78125
390625
1
0
2
0
1
0
1
0
5
0
10
0
5
0
5
0

1
8
8
2
1
0.333333
0
0.333333
0
0
0
0
0.333333
12
144
1728
20736
248832
2985984
35831808
4.3E+08
4
0
4
0
0
0
0
4
48
0
48
0
0
0
0
48

1
9
9
1
1
0
0.4
0
0.4
0.2
0
0
0
5
25
125
625
3125
15625
78125
390625
0
2
0
2
1
0
0
0
0
10
0
10
5
0
0
0

1
9
9
2
0
0.333333
0
0.333333
0
0
0
0.333333
0
12
144
1728
20736
248832
2985984
35831808
4.3E+08
4
0
4
0
0
0
4
0
48
0
48
0
0
0
48
0

1
10
10
1
1
0
0
0.4
0.4
0
0
0.2
0
5
25
125
625
3125
15625
78125
390625
0
0
2
2
0
0
1
0
0
0
10
10
0
0
5
0

1
10
10
2
0
0.4
0.4
0
0
0
0.2
0
0
5
25
125
625
3125
15625
78125
390625
2
2
0
0
0
1
0
0
10
10
0
0
0
5
0
0

1
11
11
1
1
0
0.5
0.5
0
0
0
0
0
2
4
8
16
32
64
128
256
0
1
1
0
0
0
0
0
0
2
2
0
0
0
0
0

1
11
11
2
0
0
0.4
0
0
0
0
0.6
0
10
100
1000
10000
100000
1000000
10000000
1E+08
0
4
0
0
0
0
6
0
0
40
0
0
0
0
60
0

1
12
12
1
0
0.333333
0.333333
0
0
0
0
0
0.333333
9
81
729
6561
59049
531441
4782969
43046721
3
3
0
0
0
0
0
3
27
27
0
0
0
0
0
27

1
12
12
2
1
0.166667
0
0
0.333333
0.166667
0
0
0.333333
12
144
1728
20736
248832
2985984
35831808
4.3E+08
2
0
0
4
2
0
0
4
24
0
0
48
24
0
0
48

2
13
1
1
1
0
0
0
0
0
0
1
0
1
1
1
1
1
1
1
1
0
0
0
0
0
0
1
0
0
0
0
0
0
0
1
0

2
13
1
2
0
0.333333
0.666667
0
0
0
0
0
0
3
9
27
81
243
729
2187
6561
1
2
0
0
0
0
0
0
3
6
0
0
0
0
0
0

2
14
2
1
0
0
0
1
0
0
0
0
0
1
1
1
1
1
1
1
1
0
0
1
0
0
0
0
0
0
0
1
0
0
0
0
0

2
14
2
2
1
0
0
0.25
0
0
0
0.25
0.5
4
16
64
256
1024
4096
16384
65536
0
0
1
0
0
0
1
2
0
0
4
0
0
0
4
8

I was already able to analyse a conditional logit, mixed logit and latent class model on my data. However, I also wanted to check for attribute non-attendance. As you suggested, I first tried to do this for only some parameters of my model as follow:

eaalogit chosen fpr fsec ftert mnopr mpr msec mtert, group(choicecard) id(respondent) keaa(7) eaaspec(x1 x2 x3 x4 x5 x6 x7)

This works, although after more than 600 iterations and running my command for 12h, I did not get an output yet, but after most of the iterations Stata shows that the iteration is 'not concave'.

The code for my full model (where I assume that if respondents ignored one of the main effects of the schooling components, they also would ignore the interaction term) should have to be:

eaalogit chosen fpr fsec ftert mnopr mpr msec mtert fnopr_nch fpr_nch fsec_nch ftert_nch mnopr_nch mpr_nch msec_nch mtert_nch fnopr_nch2 fpr_nch2 fsec_nch2 ftert_nch2 mnopr_nch2 mpr_nch2 msec_nch2 mtert_nch2, group(choicecard) id(respondent) keaa(23) eaaspec(x1 x2 x3 x4 x5 x6 x7 x8 x1 x2 x3 x4 x5 x6 x7 x8 x1 x2 x3 x4 x5 x6 x7)

However, after running this command for 80h, I also did not get an output yet. Do you have a suggestion what the problem could be maybe?

Many thanks in advance

Eva Boonaert

[Arne Risa Hole]

Thank you for providing more information about your dataset. The fact that the code is now running suggests that it was the misspecification issue that caused the error message.

The eaalogit model can be difficult to get to converge. One thing you can try is to specify the trace option which will display the coefficient estimates in each iteration. If one of the ANA parameters is very large after several iterations you can specify full attendance to the corresponding attribute - see the help file for an example. I can't help you more than this I'm afraid.

Arne