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: difference between -test- and -contrast- in Statat 12
From
Richard Williams <[email protected]>
To
[email protected], "[email protected]" <[email protected]>
Subject
Re: st: difference between -test- and -contrast- in Statat 12
Date
Thu, 25 Aug 2011 23:33:40 -0400
At 10:15 PM 8/25/2011, Ricardo Ovaldia wrote:
I am using the new -contrast- command in Stata 12 to test a contrast
after ANOVA and I get different p-values than when using -test-
although the marginal means are the same.
My guess is that it is because the first statistic is an F statistic
with d.f. 1, 63 while the 2nd statistic is chi-square(1). Since N is
small the P values don't match exactly. I added [fw=15] to your anova
command (to make the sample larger] and the P values were the same to
at least 4 decimal places. Try it with larger samples (or just
inflate your own sample using fweights) and see if the discrepancies disappear.
I illustrate with the auto data by first making the rep78 variable a
3 level variable and then running an ANOVA follow by -contrast-:
. sysuse auto,clear
. replace rep78=3 if rep78<3
. anova price rep78##foreign
<output omitted>
. contrast r.foreign@rep78
Contrasts of marginal linear predictions
Margins : asbalanced
-------------------------------------------------
| df F P>F
--------------+----------------------------------
foreign@rep78 |
(1 vs 0) 3 | 1 0.73 0.3965
(1 vs 0) 4 | 1 0.07 0.7881
(1 vs 0) 5 | 1 0.80 0.3743
Joint | 3 0.53 0.6606
|
Residual | 63
-------------------------------------------------
---------------------------------------------------------------
| Contrast Std. Err. [95% Conf. Interval]
--------------+------------------------------------------------
foreign@rep78 |
(1 vs 0) 3 | -1529.739 1792.049 -5110.862 2051.385
(1 vs 0) 4 | 379.8889 1407.262 -2432.3 3192.078
(1 vs 0) 5 | 2088.167 2333.681 -2575.322 6751.655
---------------------------------------------------------------
From the above tables the p-value for the test of foreign withing
rep78=3 is: p=0.3965
Now using -margin, post- follow by -test- for the same comparison
(i.e. foreign withing rep78=3):
. margins ,over(rep78 foreign) post cformat(%5.2f)
Predictive margins Number of obs = 69
Expression : Linear prediction, predict()
over : rep78 foreign
-------------------------------------------------------------------------------
| Delta-method
| Margin Std. Err. z P>|z| [95%
Conf. Interval]
--------------+----------------------------------------------------------------
rep78#foreign |
3
0 | 6358.41 490.77 12.96 0.000 5396.51 7320.30
3
1 | 4828.67 1723.54 2.80 0.005 1450.60 8206.74
4
0 | 5881.56 995.08 5.91 0.000 3931.23 7831.89
4
1 | 6261.44 995.08 6.29 0.000 4311.11 8211.77
5
0 | 4204.50 2110.89 1.99 0.046 67.22 8341.78
5
1 | 6292.67 995.08 6.32 0.000 4342.34 8243.00
-------------------------------------------------------------------------------
. test 0.foreign#3.rep78=1.foreign#3.rep78
( 1) 3bn.rep78#0bn.foreign - 3bn.rep78#1.foreign = 0
chi2( 1) = 0.73
Prob > chi2 = 0.3933
I get a different p-value. Although close in this example, in
another case that have the p-values are further apart: (0.0026 vs 0.0089).
Why are these two methods giving different answers?
Thank you,
Ricardo
Ricardo Ovaldia, MS
Statistician
Oklahoma City, OK
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/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/statalist/faq
* http://www.ats.ucla.edu/stat/stata/