跨栏泊松后估计 - 部分效应

Question

我已经运行了 Poisson Hurdle 模型，并且使用以下语法来对模型的零和计数部分产生部分效果。我目前使用的是

effects

软件包，它可能与

pscl

软件包不兼容。我也安装了

marginaleffects

软件包，但我不确定该软件包的正确功能是什么。

library(pscl)
library(effects)
data(quine, package = "MASS")

PoissonHurdle_model <- hurdle(Days ~  Eth + Sex + Age + Lrn, 
    data = quine, dist = "poisson")

partial_effects_zero <- allEffects(PoissonHurdle_model,component = "zero")

partial_effects_count <- allEffects(PoissonHurdle_model,component = "count")

我收到一条错误消息：

mod.matrix %*% scoef 中出现错误：参数不一致

我该如何解决这个问题？或者使用

marginaleffects

包计算效果的正确函数是什么？我知道我可以使用 margins 包，但它不适用于我正在使用的 R 版本。

Answer 1

我相信这些是您正在寻找的命令：

library(pscl)
library(marginaleffects)
data(quine, package = "MASS")
PoissonHurdle_model <- hurdle(
        Days ~ Eth + Sex + Age + Lrn,
        data = quine, dist = "poisson")

零分量的平均偏效应

avg_slopes(PoissonHurdle_model, type = "zero")
#;-) 
#;-)  Term            Contrast  Estimate Std. Error        z Pr(>|z|)   S   2.5 %  97.5 %
#;-)   Age mean(F1) - mean(F0)  0.000448     0.0607  0.00739   0.9941 0.0 -0.1185  0.1194
#;-)   Age mean(F2) - mean(F0)  0.016649     0.0558  0.29852   0.7653 0.4 -0.0927  0.1260
#;-)   Age mean(F3) - mean(F0) -0.003643     0.0665 -0.05479   0.9563 0.1 -0.1340  0.1267
#;-)   Eth mean(N) - mean(A)   -0.088072     0.0372 -2.36965   0.0178 5.8 -0.1609 -0.0152
#;-)   Lrn mean(SL) - mean(AL) -0.011934     0.0485 -0.24601   0.8057 0.3 -0.1070  0.0831
#;-)   Sex mean(M) - mean(F)   -0.055396     0.0418 -1.32555   0.1850 2.4 -0.1373  0.0265
#;-) 
#;-) Columns: term, contrast, estimate, std.error, statistic, p.value, s.value, conf.low, conf.high, predicted_lo, predicted_hi, predicted 
#;-) Type:  zero

响应分量的平均部分效应

avg_slopes(PoissonHurdle_model, type = "response")
#;-) 
#;-)  Term            Contrast Estimate Std. Error     z Pr(>|z|)    S   2.5 % 97.5 %
#;-)   Age mean(F1) - mean(F0)    -4.08      1.193 -3.42   <0.001 10.6  -6.416  -1.74
#;-)   Age mean(F2) - mean(F0)     4.45      1.341  3.32   <0.001 10.1   1.822   7.08
#;-)   Age mean(F3) - mean(F0)     8.08      1.811  4.46   <0.001 16.9   4.534  11.63
#;-)   Eth mean(N) - mean(A)      -8.73      0.901 -9.69   <0.001 71.3 -10.500  -6.97
#;-)   Lrn mean(SL) - mean(AL)     5.67      1.244  4.56   <0.001 17.5   3.230   8.11
#;-)   Sex mean(M) - mean(F)       2.32      0.957  2.42   0.0153  6.0   0.445   4.20
#;-) 
#;-) Columns: term, contrast, estimate, std.error, statistic, p.value, s.value, conf.low, conf.high, predicted_lo, predicted_hi, predicted 
#;-) Type:  response

部分效应，将所有协变量保持为均值或众数

slopes(PoissonHurdle_model, type = "zero", newdata = "mean")
#;-) 
#;-)  Term Contrast Estimate Std. Error       z Pr(>|z|)   S   2.5 % 97.5 %
#;-)   Age  F1 - F0  0.00104     0.0599  0.0174    0.986 0.0 -0.1163 0.1184
#;-)   Age  F2 - F0  0.01613     0.0551  0.2927    0.770 0.4 -0.0919 0.1242
#;-)   Age  F3 - F0 -0.00365     0.0654 -0.0559    0.955 0.1 -0.1318 0.1245
#;-)   Eth  N - A   -0.05161     0.0506 -1.0203    0.308 1.7 -0.1508 0.0475
#;-)   Lrn  SL - AL -0.01408     0.0520 -0.2710    0.786 0.3 -0.1159 0.0878
#;-)   Sex  M - F   -0.09043     0.0906 -0.9976    0.318 1.7 -0.2681 0.0872
#;-) 
#;-) Columns: rowid, term, contrast, estimate, std.error, statistic, p.value, s.value, conf.low, conf.high, predicted_lo, predicted_hi, predicted, Eth, Sex, Age, Lrn, Days 
#;-) Type:  zero

部分效应，将所有协变量保持为均值或众数

slopes(PoissonHurdle_model, type = "response", newdata = "mean")
#;-) 
#;-)  Term Contrast Estimate Std. Error      z Pr(>|z|)    S 2.5 % 97.5 %
#;-)   Age  F1 - F0   -2.506      0.780 -3.214  0.00131  9.6 -4.03 -0.978
#;-)   Age  F2 - F0    2.758      0.880  3.134  0.00172  9.2  1.03  4.482
#;-)   Age  F3 - F0    4.972      1.117  4.452  < 0.001 16.8  2.78  7.160
#;-)   Eth  N - A     -4.255      0.556 -7.657  < 0.001 45.6 -5.34 -3.166
#;-)   Lrn  SL - AL    2.594      0.601  4.317  < 0.001 16.0  1.42  3.772
#;-)   Sex  M - F      0.612      0.858  0.713  0.47578  1.1 -1.07  2.295
#;-) 
#;-) Columns: rowid, term, contrast, estimate, std.error, statistic, p.value, s.value, conf.low, conf.high, predicted_lo, predicted_hi, predicted, Eth, Sex, Age, Lrn, Days 
#;-) Type:  response