Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: gologit2 interaction
From
Richard Williams <[email protected]>
To
[email protected], "[email protected]" <[email protected]>
Subject
Re: st: gologit2 interaction
Date
Sun, 01 Sep 2013 18:00:19 -0500
At 04:10 PM 9/1/2013, William Buchanan wrote:
Although I thank you for following the protocol of showing your
input and output, you still have yet to acknowledge where you
downloaded the package from. In general I think many folks here
would tell you to start off with a simpler model. Is the
coefficient for the interaction term of particular interest or are
you trying to increase the model fit by including additional
parameters? How is your outcome distributed? No one other than you
can tell you whether or not you can use that in your study.
I agree with William's concerns and will add several others. First
off though, I will note that gologit2 can be downloaded from SSC and
that the support page and troubleshooting FAQs are at
http://www3.nd.edu/~rwilliam/gologit2/index.html
http://www3.nd.edu/~rwilliam/gologit2/tsfaq.html
Now, the concerns I have are
* Why are you using gologit2 in the first place? Have you tested
whether or not the assumptions of the simpler ologit model are
violated? Read up on gologit2 at the above links to learn more about
it if you haven't already. Don't go to a more complicated and harder
to understand approach until you have at least checked out whether
something simpler will do.
* Given that you are using gologit2, why aren't you using options
like autofit or pl? A totally unconstrained gologit model isn't worth
that much -- if you can't make the model more parsimonious you might
as well use mlogit, which is more widely used and understood.
* You only have 230 cases and yet you are estimating around 50
parameters. That is probably way too many for any method that uses
maximum likelihood (informal rule of thumb is at least 10 cases per
parameter). Like William says, the model probably needs to be
simplified, by dropping variables and/or by imposing proportionality
constraints.
* You say "variable like z2targetcrossus is not significant in the
raw model." How did you test that? There are actually 3 different
parameters for each variable. You either need to do a chi-square
contrast between models (i.e. estimate a model that doesn't have the
variable and then another model that does have it) or else do a wald
test, e.g. after your first gologit2 command add a command like
test z2targetcrossus
That will provide a joint test for all three of the parameters that
are estimated for that variable.
* Finally, I think you should read up a bit on interactions. You say
"variable like z2targetcrossus is not significant in the raw model,
but the interaction z2targetrd=z2targetcrossus*rd is significant. How
could this happen?" First off, we don't actually know that the z2 var
has insignificant effects; you also have the r2 variable as part of
the interaction; and in any event there is no reason that things like
this can't happen. For example, a variable could have a positive
effect on one group and a negative effect on the other. The main
effect alone might come out as zero or thereabouts but once you split
the effects out for each group the sign and significance of the
effects could become clear. A possible starting place for more
reading on interactions (I am not saying it is the best, but it is
one I can find in 10 seconds) is
http://www3.nd.edu/~rwilliam/stats2/l53.pdf
Hope this helps.
HTH,
Billy
Sent from my iPhone
On Sep 1, 2013, at 14:59, lan zhang <[email protected]> wrote:
> the raw model:
> gologit2 ma2 z2targetcrossus z2usinl z2chcross z2chdo woe jv rd
network repoffice z2age unique z2location z2chsenior z2reve
>> nue z2mconcentration z2var65
>
> Generalized Ordered Logit Estimates Number of
obs = 230
>
LR chi2(48) = 131.44
>
Prob > chi2 = 0.0000
> Log likelihood = -190.91867 Pseudo
R2 = 0.2561
>
>
----------------------------------------------------------------------------------
> ma2 | Coef. Std. Err. z P>|z| [95%
Conf. Interval]
>
-----------------+----------------------------------------------------------------
> 0 |
> z2targetcrossus
| -.0173887 .1668386 -0.10 0.917 -.3443863 .3096089
> z2usinl
| -.2492047 .1905918 -1.31 0.191 -.6227578 .1243484
> z2chcross
| .103573 .1520177 0.68 0.496 -.1943763 .4015223
> z2chdo
| -.0393117 .1750591 -0.22 0.822 -.3824213 .3037979
> woe
| -.4751755 .6626994 -0.72 0.473 -1.774043 .8236915
> jv
| -1.083191 .7963511 -1.36 0.174 -2.644011 .4776286
> rd
| .9039168 .4271417 2.12 0.034 .0667344 1.741099
> network
| 1.410567 .9212801 1.53 0.126 -.3951092 3.216242
> repoffice
| -1.001058 .4313348 -2.32 0.020 -1.846458 -.1556568
> z2age
| .1168134 .1774538 0.66 0.510 -.2309898 .4646165
> unique
| .6194533 .3772757 1.64 0.101 -.1199934 1.3589
> z2location
| .1114399 .1717018 0.65 0.516 -.2250895 .4479693
> z2chsenior
| .0730508 .1674231 0.44 0.663 -.2550924 .401194
> z2revenue
| -.014403 .165149 -0.09 0.931 -.3380891 .3092831
> z2mconcentration
| -.2459377 .1718248 -1.43 0.152 -.5827081 .0908327
> z2var65
| .2098376 .1512144 1.39 0.165 -.0865371 .5062123
> _cons
| -.4365704 .6758439 -0.65 0.518 -1.7612 .8880593
>
-----------------+----------------------------------------------------------------
> 1 |
> z2targetcrossus
| .5896636 .3993497 1.48 0.140 -.1930474 1.372375
> z2usinl
| -5.757507 3.010157 -1.91 0.056 -11.65731 .1422915
> z2chcross
| .3443698 .8501671 0.41 0.685 -1.321927 2.010667
> z2chdo
| .0274867 .3531697 0.08 0.938 -.6647132 .7196866
> woe
| -.1374271 .8539173 -0.16 0.872 -1.811074 1.53622
> jv
| -1.625354 1.073093 -1.51 0.130 -3.728578 .47787
> rd
| .2289301 .5789769 0.40 0.693 -.9058436 1.363704
> network
| .9461952 1.146079 0.83 0.409 -1.300079 3.192469
> repoffice
| .1904672 .6567114 0.29 0.772 -1.096663 1.477598
> z2age
| .3646071 .2886476 1.26 0.207 -.2011317 .930346
> unique
| .1462646 .6053367 0.24 0.809 -1.040174 1.332703
> z2location
| .3362546 .2857039 1.18 0.239 -.2237147 .896224
> z2chsenior
| -.6377379 .2899929 -2.20 0.028 -1.206114 -.0693622
> z2revenue
| .4384878 .2596045 1.69 0.091 -.0703277 .9473032
> z2mconcentration
| -.3356207 .3044584 -1.10 0.270 -.9323481 .2611067
> z2var65
| .680389 .2555405 2.66 0.008 .1795389 1.181239
> _cons
| -3.852209 1.534456 -2.51 0.012 -6.859688 -.8447304
>
-----------------+----------------------------------------------------------------
> 2 |
> z2targetcrossus
| -1.859479 .9033436 -2.06 0.040 -3.63 -.0889579
> z2usinl
| 29.60616 10.01538 2.96 0.003 9.97637 49.23595
> z2chcross
| -6.148175 2.31656 -2.65 0.008 -10.68855 -1.607802
> z2chdo
| -2.199313 .9065691 -2.43 0.015 -3.976156 -.4224705
> woe
| 4.507415 2.227588 2.02 0.043 .1414227 8.873408
> jv
| 20.13766 566.9975 0.04 0.972 -1091.157 1131.432
> rd
| -3.677817 1.353729 -2.72 0.007 -6.331076 -1.024557
> network
| -20.33935 566.9957 -0.04 0.971 -1131.631 1090.952
> repoffice
| 5.012434 2.120188 2.36 0.018 .8569423 9.167927
> z2age
| .5531285 .5823306 0.95 0.342 -.5882184 1.694475
> unique
| -7.34904 2.068999 -3.55 0.000 -11.4042 -3.293877
> z2location
| 3.362253 .9824052 3.42 0.001 1.436774 5.287732
> z2chsenior
| .4177627 .4609106 0.91 0.365 -.4856056 1.321131
> z2revenue
| .8860638 .4915586 1.80 0.071 -.0773735 1.849501
> z2mconcentration
| 1.555764 .692565 2.25 0.025 .1983613 2.913166
> z2var65
| -.5325391 .556961 -0.96 0.339 -1.624163 .5590843
> _cons
| 8.84202 3.903878 2.26 0.024 1.19056 16.49348
>
----------------------------------------------------------------------------------
> model with interaction:
> . gologit2 ma2 z2targetcrossus z2usinl z2chcross z2chdo woe jv rd
network repoffice z2age unique z2location z2chsenior z2reve
>> nue z2mconcentration z2var65 z2targetrd
>
> Generalized Ordered Logit Estimates Number of
obs = 230
> LR
chi2(51) = 148.17
> Prob >
chi2 = 0.0000
> Log likelihood = -182.55401 Pseudo
R2 = 0.2887
>
>
----------------------------------------------------------------------------------
> ma2 | Coef. Std. Err. z P>|z| [95%
Conf. Interval]
>
-----------------+----------------------------------------------------------------
> 0 |
> z2targetcrossus
| -.1086327 .1786952 -0.61 0.543 -.4588687 .2416034
> z2usinl
| -.224747 .1906251 -1.18 0.238 -.5983653 .1488713
> z2chcross
| .1182669 .153385 0.77 0.441 -.1823621 .418896
> z2chdo
| -.1046159 .1873578 -0.56 0.577 -.4718305 .2625987
> woe
| -.531372 .6963073 -0.76 0.445 -1.896109 .8333653
> jv
| -1.16429 .8444907 -1.38 0.168 -2.819462 .490881
> rd
| -.2352473 .6686808 -0.35 0.725 -1.545838 1.075343
> network
| 1.654026 .9829236 1.68 0.092 -.2724686 3.580521
> repoffice
| -1.165368 .4452497 -2.62 0.009 -2.038042 -.2926951
> z2age
| .0913011 .1810404 0.50 0.614 -.2635316 .4461339
> unique
| .6011205 .3888613 1.55 0.122 -.1610336 1.363275
> z2location
| .1073559 .1764745 0.61 0.543 -.2385278 .4532396
> z2chsenior
| .111229 .1726965 0.64 0.520 -.2272498 .4497079
> z2revenue
| -.0261089 .1675079 -0.16 0.876 -.3544183 .3022005
> z2mconcentration
| -.2419455 .176234 -1.37 0.170 -.5873579 .1034668
> z2var65
| .2004429 .1538167 1.30 0.193 -.1010322 .5019181
> z2targetrd
| .5360898 .2728751 1.96 0.049 .0012645 1.070915
> _cons
| -.1930397 .7108041 -0.27 0.786 -1.58619 1.200111
>
-----------------+----------------------------------------------------------------
> 1 |
> z2targetcrossus
| .9581676 .4375794 2.19 0.029 .1005277 1.815808
> z2usinl
| -7.478695 3.056712 -2.45 0.014 -13.46974 -1.48765
> z2chcross
| .183797 1.136537 0.16 0.872 -2.043775 2.411369
> z2chdo
| .0680997 .3728366 0.18 0.855 -.6626467 .7988461
> woe
| -.0332427 .8441079 -0.04 0.969 -1.687664 1.621178
> jv
| -2.196873 1.394514 -1.58 0.115 -4.930071 .5363247
> rd
| 2.356204 1.012723 2.33 0.020 .3713034 4.341105
> network
| .5403172 1.4498 0.37 0.709 -2.301238 3.381872
> repoffice
| 1.179 .7543806 1.56 0.118 -.2995585 2.657559
> z2age
| .3700746 .3043611 1.22 0.224 -.2264622 .9666115
> unique
| .0186732 .6888141 0.03 0.978 -1.331378 1.368724
> z2location
| .4302243 .3273268 1.31 0.189 -.2113244 1.071773
> z2chsenior
| -.9859917 .3561425 -2.77 0.006 -1.684018 -.2879651
> z2revenue
| .5751533 .2707602 2.12 0.034 .0444731 1.105833
> z2mconcentration
| -.287354 .334713 -0.86 0.391 -.9433795 .3686715
> z2var65
| .7992744 .2936887 2.72 0.006 .2236551 1.374894
> z2targetrd
| -.7707145 .3850767 -2.00 0.045 -1.525451 -.0159781
> _cons
| -4.906128 1.712643 -2.86 0.004 -8.262847 -1.54941
>
-----------------+----------------------------------------------------------------
> 2 |
> z2targetcrossus
| -1.40224 .940126 -1.49 0.136 -3.244853 .440373
> z2usinl
| 27.85302 9.856849 2.83 0.005 8.533951 47.17209
> z2chcross
| -5.887844 2.251756 -2.61 0.009 -10.3012 -1.474484
> z2chdo
| -2.131811 .9187053 -2.32 0.020 -3.93244 -.3311814
> woe
| 4.412064 2.322375 1.90 0.057 -.1397075 8.963836
> jv
| 19.98229 628.5635 0.03 0.975 -1211.98 1251.944
> rd
| -2.206978 1.697812 -1.30 0.194 -5.534629 1.120673
> network
| -20.38483 628.5619 -0.03 0.974 -1252.343 1211.574
> repoffice
| 5.196042 2.137634 2.43 0.015 1.006356 9.385727
> z2age
| .7685141 .6105997 1.26 0.208 -.4282392 1.965267
> unique
| -8.068043 2.163387 -3.73 0.000 -12.3082 -3.827882
> z2location
| 3.371957 1.005542 3.35 0.001 1.401131 5.342782
> z2chsenior
| .3486575 .5132131 0.68 0.497 -.6572217 1.354537
> z2revenue
| .9434554 .5230482 1.80 0.071 -.0817001 1.968611
> z2mconcentration
| 1.711886 .6913372 2.48 0.013 .3568901 3.066882
> z2var65
| -.4395993 .5783425 -0.76 0.447 -1.57313 .6939311
> z2targetrd
| -.6754823 .5818942 -1.16 0.246 -1.815974 .4650093
> _cons
| 8.627572 3.824513 2.26 0.024 1.131665 16.12348
>
----------------------------------------------------------------------------------
>
> my question is: variable like z2targetcrossus is not significant
in the raw model, but the interaction z2targetrd=z2targetcrossus*rd
is significant.
> How could this happen?
> can i still use this result in my study?
>
> *
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/faqs/resources/statalist-faq/
> * http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME: (574)289-5227
EMAIL: [email protected]
WWW: http://www.nd.edu/~rwilliam
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/