3
我正在R（http://www.ats.ucla.edu/stat/r/dae/zinbreg.htm）和Stata（http://www.ats.ucla.edu/stat/stata/dae/zinb.htm）上运行带有probit链接的零膨胀负二项模型。Vuong测试在R和Stata上有不同的结果
有一个Vuong测试来比较这个规范是否比普通的负二项模型更好。 R告诉我使用后者更好，Stata说ZINB是更好的选择。在这两种情况下，我都假设导致过零点的过程与负二项分布式非零观测值相同。系数确实是一样的（除了Stata打印一位数字以外）。
在R我运行（数据代码是下面）
require(pscl)
ZINB <- zeroinfl(Two.Year ~ length + numAuth + numAck,
data=Master,
dist="negbin", link="probit"
)
NB <- glm.nb(Two.Year ~ length + numAuth + numAck,
data=Master
)
均与vuong(ZINB, NB)从同一封装的产率
Vuong Non-Nested Hypothesis Test-Statistic: -10.78337
(test-statistic is asymptotically distributed N(0,1) under the
null that the models are indistinguishible)
in this case:
model2 > model1, with p-value < 2.22e-16
因此比较：NB比ZINB更好。
在Stata我跑
zinb twoyear numauth length numack, inflate(numauth length numack) probit vuong
和接收（迭代拟合抑制）
Zero-inflated negative binomial regression Number of obs = 714
Nonzero obs = 433
Zero obs = 281
Inflation model = probit LR chi2(3) = 74.19
Log likelihood = -1484.763 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
twoyear | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
twoyear |
numauth | .1463257 .0667629 2.19 0.028 .0154729 .2771785
length | .038699 .006077 6.37 0.000 .0267883 .0506097
numack | .0333765 .010802 3.09 0.002 .0122049 .0545481
_cons | -.4588568 .2068824 -2.22 0.027 -.8643389 -.0533747
-------------+----------------------------------------------------------------
inflate |
numauth | .2670777 .1141893 2.34 0.019 .0432708 .4908846
length | .0147993 .0105611 1.40 0.161 -.0059001 .0354987
numack | .0177504 .0150118 1.18 0.237 -.0116722 .0471729
_cons | -2.057536 .5499852 -3.74 0.000 -3.135487 -.9795845
-------------+----------------------------------------------------------------
/lnalpha | .0871077 .1608448 0.54 0.588 -.2281424 .4023577
-------------+----------------------------------------------------------------
alpha | 1.091014 .175484 .7960109 1.495346
------------------------------------------------------------------------------
Vuong test of zinb vs. standard negative binomial: z = 2.36 Pr>z = 0.0092
在最后一行塔塔告诉我，在这种情况下ZINB比NB好：两个检验统计量和p值不同。怎么来的？
数据（R代码）
Master <- <-read.table(text="
Two.Year numAuth length numAck
0 1 4 6
3 3 28 3
3 1 18 4
0 1 42 4
0 2 17 0
2 1 10 3
1 2 20 0
0 1 28 3
1 1 23 7
0 2 34 3
2 2 24 2
0 2 18 0
0 1 23 7
0 1 35 11
4 2 33 13
0 2 24 4
0 2 21 9
1 4 21 0
1 1 8 6
2 1 18 1
0 3 28 2
0 2 17 2
1 1 30 6
4 2 28 16
1 4 35 1
2 3 19 2
0 1 24 2
1 3 26 6
1 1 17 7
0 3 42 4
0 3 32 8
3 1 33 23
7 2 24 9
0 2 25 6
1 1 7 1
0 1 15 2
2 2 16 2
0 1 23 6
2 3 18 7
0 1 28 5
0 1 12 2
1 1 25 4
0 4 18 1
1 2 32 6
1 1 15 2
2 2 14 4
0 2 24 9
0 3 30 9
0 2 19 9
0 2 14 2
2 2 23 3
0 2 18 0
1 3 13 4
0 1 10 4
0 1 24 8
0 2 22 9
2 3 29 5
2 1 25 5
0 2 17 4
1 2 24 0
0 2 26 0
2 2 33 12
1 4 17 2
1 1 25 8
3 1 36 11
0 1 10 4
9 2 60 22
0 2 18 3
2 3 19 6
2 2 23 7
2 2 26 0
1 1 20 5
4 2 31 4
0 2 21 2
0 1 24 12
1 1 12 1
1 3 26 5
1 4 32 8
2 3 21 1
3 3 26 3
4 2 36 6
3 3 28 2
1 3 27 1
0 2 12 5
0 3 24 4
0 2 35 1
0 2 17 2
3 2 28 3
0 3 29 8
0 2 20 3
3 2 28 0
11 1 30 2
0 3 22 2
21 3 59 24
0 2 15 5
0 2 22 2
5 4 33 0
0 2 21 2
4 2 21 0
0 3 25 9
2 2 31 5
1 2 23 1
2 3 25 0
0 1 13 3
0 1 22 7
0 1 16 3
6 1 18 4
2 2 19 7
3 2 22 10
0 1 12 6
0 1 23 8
1 1 23 9
1 2 32 15
1 3 26 8
1 3 15 2
0 3 16 2
0 4 29 2
2 3 24 3
2 3 32 1
2 1 29 13
1 3 26 0
5 1 23 4
3 2 21 2
4 2 19 4
4 3 19 2
2 1 29 0
0 1 13 6
0 2 28 2
0 3 33 1
0 1 20 2
0 1 30 8
1 2 19 2
17 2 30 7
5 3 39 17
21 3 30 5
1 3 29 24
1 1 31 4
4 3 26 13
4 2 14 16
2 3 31 14
5 3 37 10
15 2 52 13
1 1 6 5
2 1 24 13
17 3 17 3
3 2 29 5
2 1 26 7
3 3 34 9
5 2 39 2
3 1 26 7
1 2 32 12
2 3 26 4
9 3 28 8
1 3 29 1
4 1 24 7
9 1 40 13
1 2 27 21
2 2 27 13
5 3 31 10
10 2 29 15
10 2 41 15
8 1 24 17
2 4 16 5
17 2 26 20
3 2 31 3
2 2 18 1
6 3 32 9
2 1 32 11
4 3 34 8
4 1 16 1
5 1 33 5
0 2 17 11
17 2 48 8
2 1 11 2
5 3 33 18
4 2 25 9
10 2 17 5
1 1 25 8
3 3 41 16
2 1 40 13
4 3 25 2
16 4 32 13
10 1 33 18
5 2 25 3
3 2 20 3
2 3 14 7
3 2 23 4
2 2 28 4
3 2 25 19
0 2 14 6
3 1 28 18
8 3 27 11
1 3 25 17
21 2 33 15
9 2 24 2
1 1 16 14
1 1 38 10
16 2 37 13
16 2 41 1
7 2 24 18
4 2 17 5
4 1 37 32
3 1 37 8
13 2 35 6
15 1 23 11
7 1 47 11
3 1 16 6
12 2 36 6
7 1 24 17
4 2 24 8
14 2 24 9
15 2 24 11
0 3 19 4
0 4 28 9
1 1 5 3
11 1 28 15
5 1 33 5
10 2 21 9
3 3 28 8
2 3 13 2
11 2 41 8
4 2 24 11
3 1 32 11
4 2 31 11
7 2 34 3
11 6 33 6
7 3 33 7
2 2 37 13
7 3 19 9
1 2 14 3
6 2 15 11
11 3 37 12
0 2 20 5
7 4 13 6
17 1 52 14
9 3 47 30
1 2 32 27
30 3 36 19
2 2 12 5
3 1 30 7
4 2 19 11
32 3 45 14
13 1 17 7
16 2 24 4
5 1 32 13
7 3 29 14
5 2 46 2
1 2 21 6
1 3 13 17
11 1 41 16
6 2 33 1
7 1 31 20
0 1 16 13
6 3 26 8
11 2 46 7
8 2 20 5
8 1 44 7
2 2 33 12
1 3 22 5
0 4 14 2
4 1 25 8
5 3 24 11
1 1 21 18
5 1 28 5
2 1 51 19
2 1 16 4
17 2 35 2
4 1 35 1
9 3 48 8
2 1 33 16
0 3 24 7
18 2 33 12
11 1 41 5
5 2 17 3
8 1 19 7
4 3 38 2
23 2 27 10
22 3 46 13
5 3 21 1
5 2 38 10
1 2 20 5
2 2 24 8
0 3 30 9
7 2 44 16
7 1 21 7
0 1 20 10
10 2 33 11
4 2 18 2
11 1 45 17
7 2 32 7
7 2 28 6
5 2 25 10
3 2 57 6
8 1 16 2
7 2 34 4
5 2 22 8
2 2 21 7
4 2 37 15
2 4 36 7
1 1 17 4
0 2 23 9
12 2 48 4
8 3 29 13
0 1 29 7
0 2 27 12
1 1 53 10
3 3 15 5
8 1 40 29
2 2 22 11
10 2 20 7
4 4 27 3
4 1 24 4
2 2 24 5
1 2 19 6
10 3 41 10
57 3 46 9
5 1 20 11
6 2 30 4
0 2 20 5
16 3 35 8
1 2 44 1
2 4 24 8
1 1 20 9
5 3 19 11
5 3 29 15
3 1 21 8
3 3 19 3
8 3 44 0
11 3 34 15
2 2 31 1
11 1 39 11
0 3 24 3
4 2 35 6
2 1 14 6
10 1 30 10
6 2 21 4
9 2 32 3
0 1 34 10
6 2 32 3
7 2 50 11
11 1 35 15
4 1 27 9
1 2 32 27
8 2 54 2
0 3 15 8
2 1 31 13
0 1 31 11
0 4 14 5
0 2 37 15
0 2 51 12
0 2 34 1
0 3 29 12
0 2 22 11
0 2 19 15
0 2 39 13
0 3 25 12
0 1 46 2
0 4 42 10
0 1 38 5
0 3 31 4
0 3 33 1
0 2 24 11
0 1 28 16
0 2 28 13
0 1 29 17
0 1 23 13
0 3 36 21
0 2 30 15
0 2 25 12
0 2 26 17
0 3 19 2
0 2 37 5
0 2 47 12
0 1 21 20
0 3 27 21
0 2 16 7
0 1 35 5
0 2 32 24
0 3 31 6
0 3 36 13
0 2 26 20
0 1 31 13
0 2 46 6
0 2 34 12
0 1 18 13
0 1 29 3
0 3 40 9
0 1 25 3
0 3 45 9
0 2 31 3
0 2 35 4
0 3 29 10
0 2 33 13
0 3 22 4
0 2 26 9
0 2 29 19
0 2 28 12
0 2 30 5
0 4 30 3
0 3 32 14
0 3 45 20
0 2 42 9
0 2 25 4
0 2 20 22
0 3 31 5
0 1 26 13
0 2 32 11
0 1 31 2
0 2 42 17
0 1 37 8
0 3 37 16
0 3 25 10
0 2 33 11
0 2 29 7
0 2 21 16
0 3 30 33
0 1 35 8
0 3 25 6
0 2 54 3
0 2 41 10
0 3 35 1
0 4 26 4
0 2 31 4
0 3 26 11
0 3 34 11
0 2 27 7
0 1 19 14
0 1 38 9
0 2 24 1
0 3 30 20
0 4 43 13
0 2 20 10
0 2 38 1
0 2 41 6
0 1 20 9
0 2 34 2
0 2 24 5
0 2 24 2
0 1 31 19
0 3 49 7
0 1 26 0
0 2 44 6
0 3 36 13
0 3 31 14
0 2 30 20
0 1 27 13
0 2 28 9
0 2 22 20
0 4 36 34
0 3 25 3
0 2 29 17
0 2 40 8
0 2 39 17
0 4 29 8
0 1 27 22
0 1 21 10
0 3 17 5
0 3 28 10
0 1 27 7
0 3 40 7
0 2 21 4
0 1 33 14
0 1 31 14
0 3 37 13
0 2 23 9
0 2 25 1
0 2 30 1
0 2 30 12
0 1 41 8
0 2 26 1
0 2 25 14
0 2 26 3
0 3 36 1
0 4 23 1
0 2 18 0
0 2 34 2
0 1 39 6
0 1 16 15
0 3 34 4
0 4 35 6
0 1 22 10
0 1 35 8
0 2 36 13
0 2 50 8
0 2 28 6
0 1 30 14
0 2 33 26
0 3 28 1
0 1 18 10
0 2 27 4
0 2 27 5
0 2 8 2
0 4 32 16
0 3 40 6
0 4 45 15
0 2 38 3
0 2 29 6
0 1 25 9
12 1 27 5
2 1 33 8
4 3 31 3
1 1 33 4
0 3 20 5
0 2 28 6
2 2 32 12
0 3 30 2
0 3 19 3
1 1 14 19
0 2 28 2
0 3 26 3
0 2 32 13
1 3 21 7
1 4 20 0
2 2 40 8
0 2 35 18
1 1 20 6
6 2 21 3
3 2 33 10
1 1 31 15
1 2 22 5
0 2 24 7
2 2 22 3
3 2 17 6
9 2 30 12
2 4 39 9
0 2 46 8
0 2 26 5
1 2 28 5
6 1 18 3
5 2 19 13
1 3 27 3
1 1 20 10
0 1 27 6
0 4 26 1
0 2 19 4
0 1 26 8
1 1 30 8
0 2 22 2
3 3 42 4
3 1 10 5
3 1 30 12
1 1 25 8
1 2 38 8
2 1 28 13
3 1 18 12
2 2 20 11
2 2 29 0
1 2 18 3
1 1 6 2
0 1 6 3
2 2 24 1
0 1 14 1
1 1 17 5
2 2 20 9
1 4 24 0
1 2 8 10
0 2 18 1
1 1 25 5
2 2 12 7
0 3 18 1
0 1 19 1
8 2 21 2
1 2 23 5
7 2 19 6
1 1 21 5
0 1 16 6
1 1 24 1
0 2 19 3
1 2 14 6
3 2 24 2
6 1 32 21
0 1 16 0
1 2 15 0
1 2 8 8
0 1 14 5
0 2 27 5
2 2 17 2
1 1 19 7
1 2 21 2
0 1 29 7
0 2 18 2
0 2 15 6
2 3 27 3
0 2 57 4
2 3 17 2
1 1 18 8
1 1 17 5
0 1 18 1
1 2 18 4
1 1 12 1
0 2 15 6
1 2 24 4
3 2 14 9
0 1 24 6
3 1 30 9
0 1 19 5
3 1 16 7
5 3 21 1
2 2 17 5
4 1 34 9
1 1 17 7
3 2 30 10
12 1 17 6
2 1 26 6
1 1 18 2
2 2 24 0
0 1 12 2
0 2 3 2
1 1 11 4
1 4 18 13
0 1 25 9
8 2 20 7
0 1 11 7
7 3 26 19
6 1 18 6
6 2 32 5
1 1 31 2
1 2 33 9
4 1 17 6
1 2 34 11
5 1 37 3
0 3 27 10
12 2 25 14
3 1 40 6
6 2 27 9
0 2 31 2
1 1 28 7
2 1 37 11
1 1 19 0
5 2 30 17
4 3 40 6
0 1 27 6
5 3 31 7
0 3 26 10
3 2 32 4
1 3 43 6
3 1 19 3
2 2 37 4
0 3 28 4
6 3 30 11
1 1 30 9
4 3 31 26
1 2 14 1
10 1 35 27
1 1 36 7
5 1 32 8
2 1 28 6
3 1 34 16
3 2 32 5
1 3 11 0
2 2 42 5
0 2 30 7
0 1 32 9
3 3 43 2
7 2 43 6
1 2 21 5
2 1 27 20
1 2 37 7
2 1 37 8
0 1 19 3
0 3 28 5
2 2 33 3
3 1 41 6
13 2 41 9
2 1 38 3
4 1 32 5
2 1 34 8
1 1 27 9
8 1 29 7
4 1 17 6
0 1 20 8
1 2 34 4
1 1 16 11
4 2 33 5
0 2 15 6
1 1 27 4
2 3 15 8
1 1 30 8
3 2 41 20
0 1 25 15
1 3 35 24
4 2 30 21
6 2 30 6
16 2 33 21
2 3 37 3
2 2 30 12
4 1 57 11
0 2 18 16
4 4 20 13
3 1 43 10
3 1 25 15
7 2 31 11
2 1 31 3
5 2 40 11
3 2 28 7
4 2 27 10
0 1 26 6
4 2 24 14
4 2 23 8
0 2 25 11
21 2 33 12
1 3 37 0
3 2 28 7
4 2 27 10
1 2 41 15
2 2 30 16
2 2 28 7
6 1 19 8
4 4 22 19
0 2 38 33
1 1 29 11
1 2 27 2
4 2 24 6
2 1 22 5
",header=TRUE,sep="")
我想你可以很容易地计算Vuong测试，一旦你估计了两个模型。请参阅Greene的配方书。你也可以用'library（pscl）'看到它是如何在R中计算的，然后输入'vuong'，在Stata中你可以输入'viewsource zinb.ado'。 – Metrics 2014-09-13 14:36:24