is this a bug? [was Re: again with polr]

Sat, Mar 15, 2008 7:21 AM

Hello everybody

as I said, i'm not a ordered probit guru, but what happened seems
to me a little strange.

Here I put my dataset

http://bragliozzo.altervista.org/asd.dta

this is the code you can try

# -------------------
library(foreign)
library(MASS)
asd <- read.dta("asd.dta",
                convert.underscore = F)

base.op <- polr(
                as.ordered(lz2_1) ~
                ly +
                lnfiglimin +
                lnfiglimagg +
                lnadulti +
                eta +
                eta2 +
                uomo +
                laurea +
                saluteno +
                extra
                ,
                ##                 weights = w,
                data = asd,
                method = "probit",
                Hess = T
                )

summary(base.op)

# -------------------

The problem is: you can try _any_ combination of the explaining
variables, but when you insert eta2 (eta == age in years, eta2 ==
age^2) you can't do the summary. I don't see why: it's a variable
like others with no strange problems (at least it seems to me to
be so).

If I'm wrong please tell me, otherwise I'll submit a bug report

ok

On 15/03/08 - 10:29, Luca Braglia wrote:

hello everybody

solved the problem with summary, now I have another one

eg I estimate

try.op <- polr(
               as.ordered(sod.sit.ec.fam) ~
               log(y) +
               log(1 + nfiglimin) +
               log(1 + nfiglimagg) +
               log(ncomp - nfiglitot) +

               eta +
               I(eta^2) +
               uomo +
               elem +
               laurea +
               saluteno +
               extra
               ,

               data = dfscale,
               method = "probit",
               Hess = T
               )


now I've putted Hess=T in order to perform a summary on the
estimate. I missed this option last time

try.op
Call:
polr(formula = as.ordered(sod.sit.ec.fam) ~ log(y) + log(1 + 
    nfiglimin) + log(1 + nfiglimagg) + log(ncomp - nfiglitot) + 
    eta + I(eta^2) + uomo + elem + laurea + saluteno + extra, 
    data = dfscale, Hess = T, method = "probit")

Coefficients:
                log(y)     log(1 + nfiglimin) 
            0.76200482            -0.17791762 
   log(1 + nfiglimagg) log(ncomp - nfiglitot) 
           -0.33306775            -0.22487707 
                   eta               I(eta^2) 
           -0.03177126             0.00030204 
           uomomaschio                   elem 
           -0.02737021             0.05259243 
                laurea               saluteno 
            0.19143184            -0.23067414 
                 extra 
           -0.27810954

Intercepts:
    0|1     1|2     2|3     3|4     4|5     5|6     6|7     7|8 
5.22647 5.42504 5.76277 6.03395 6.31709 6.88459 7.27907 7.82273 
    8|9    9|10 
8.50723 8.83105

Residual Deviance: 16325.59 
AIC: 16367.59 
(58 observations deleted due to missingness)


unfortunately when I do

summary(try.op)

I get

summary(try.op)
Errore in svd(X) : valori infiniti o mancanti in 'x'

... infinite  or missing values. 

If I exclude I(eta^2) from the estimation I can do summaries: but
I need also I(eta^2) in order to compare this model with the same
specification estimated with Ols

Can you please help me or give me an hint?

If it can be useful, i hereby put an estimation with stata's
oprobit function (varnames are a little bit different)

. oprobit lz2_1 ly lnfiglimin lnfiglimagg lnadulti eta eta2 uomo elem laurea saluteno extra

Iteration 0:   log likelihood = -7877.9258
Iteration 1:   log likelihood = -7633.4001
Iteration 2:   log likelihood = -7633.3455

Ordered probit estimates                          Number of obs   =       3747
                                                  LR chi2(11)     =     489.16
                                                  Prob > chi2     =     0.0000
Log likelihood = -7633.3455                       Pseudo R2       =     0.0310

------------------------------------------------------------------------------
       lz2_1 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
          ly |   .6998939   .0369605    18.94   0.000     .6274526    .7723351
  lnfiglimin |  -.1177089   .0467595    -2.52   0.012     -.209356   -.0260619
 lnfiglimagg |  -.3195756   .0475709    -6.72   0.000    -.4128128   -.2263383
    lnadulti |  -.2644527   .0588268    -4.50   0.000     -.379751   -.1491544
         eta |  -.0245736   .0048287    -5.09   0.000    -.0340376   -.0151096
        eta2 |    .000236   .0000485     4.87   0.000     .0001409     .000331
        uomo |  -.0226491     .03348    -0.68   0.499    -.0882686    .0429705
        elem |  -.0072757   .0524754    -0.14   0.890    -.1101257    .0955743
      laurea |    .119691   .0527058     2.27   0.023     .0163895    .2229925
    saluteno |  -.1299579   .0724971    -1.79   0.073    -.2720495    .0121337
       extra |  -.3077715   .0963483    -3.19   0.001    -.4966107   -.1189323
-------------+----------------------------------------------------------------
       _cut1 |   4.403418   .3797857          (Ancillary parameters)
       _cut2 |   5.018157   .3793925 
       _cut3 |   5.372828   .3796109 
       _cut4 |   5.705886   .3800993 
       _cut5 |   6.314861   .3814847 
       _cut6 |   6.723659   .3826026 
       _cut7 |   7.278588   .3842792 
       _cut8 |   7.963981    .386254 
       _cut9 |   8.288727   .3871242 
------------------------------------------------------------------------------

. log close
       log:  z:/home/luca/stata_oprob.log
  log type:  text
 closed on:  15 Mar 2008, 10:13:50
--------------------------------------------------------------------------------------------------------



thank you
   Luca

-- 
             Luca Braglia, aka Bragliozzo
           http://bragliozzo.altervista.org

Luca Braglia, aka Bragliozzo
           http://bragliozzo.altervista.org

is this a bug? [was Re: again with polr]

Thread (3 messages)