如何从完全模型平均系数(收缩)获得参数估计的标准误差

时间:2015-02-27 02:00:37

标签: r standard-error model average

我使用MuMIn来计算模型平均过程中的参数估计值。现在我想比较条件与收缩的参数估计。我想比较两种方法的参数估计值和标准误差,但我遇到了问题。我运行下面的代码来获得一个总结表,看起来很棒。我想要摘要表中的内容。但是当我尝试从汇总表中选择收缩表时,它只给出了参数估计值,而不是完整的表格。

我的数据样本为here

这是我的代码。它改编自先前版本的一系列因变量。

data <- read.csv(sample)
xvar <- colnames(data)[2:11]
yvar <- colnames(data)[1]

formlist <- list()
for(i in yvar){
    formlist[[paste(i)]] <- as.formula(paste(i, "~ ", paste(xvar, collapse="+")))
}

library("MuMIn")

    options(na.action = "na.fail")
    name  <- names(formlist)[1]
    fm1   <- lm(formlist[[1]], data=data)
    ms1   <- dredge(fm1)
    modav <- model.avg(ms1, subset = delta < 4)
    options(na.action = "na.pass")
    modname   <- paste0("Y", rep(1, 10))
    relimp    <- summary(modav)$importance[1:10]
    parameter <- as.data.frame(cbind(names(relimp), modname))
    names(parameter) <- c("Parameter", "Dependent.var")
    parest <- coefficients(modav)[2:11]
    antm[[paste(name)]] <- cbind(parameter,relimp, parest)
    coeff  <- modav[3]
    shrink <- modav[4]
}

我被困在最后两行。如果我运行summary(modav),我会得到下表:

Call:
model.avg.model.selection(object = ms1, subset = delta < 4)

Component models: 
           df  logLik    AICc Delta Weight
4/5         4 -800.73 1609.63  0.00   0.04
5           3 -802.12 1610.34  0.70   0.03
3/4/5       5 -800.10 1610.45  0.82   0.03
4/5/9       5 -800.21 1610.68  1.05   0.03
4/5/10      5 -800.41 1611.07  1.44   0.02
2/5         4 -801.47 1611.11  1.48   0.02
4           3 -802.55 1611.21  1.57   0.02
1/4/5       5 -800.48 1611.22  1.59   0.02
4/9         4 -801.54 1611.25  1.62   0.02
5/6         4 -801.58 1611.34  1.70   0.02
3/4         4 -801.62 1611.42  1.78   0.02
5/10        4 -801.63 1611.44  1.80   0.02
4/5/6       5 -800.61 1611.47  1.84   0.02
3/4/5/10    6 -799.58 1611.52  1.88   0.02
2/4/5       5 -800.67 1611.60  1.97   0.02
4/5/8       5 -800.67 1611.60  1.97   0.02
4/5/7       5 -800.73 1611.72  2.08   0.02
5/9         4 -801.79 1611.74  2.11   0.01
1/3/4/5     6 -799.77 1611.90  2.27   0.01
3/4/10      5 -800.87 1612.00  2.37   0.01
3/4/5/9     6 -799.87 1612.10  2.47   0.01
1/5         4 -801.99 1612.14  2.51   0.01
3/4/5/6     6 -799.90 1612.15  2.52   0.01
1/4/5/9     6 -799.90 1612.16  2.53   0.01
4/5/9/10    6 -799.94 1612.24  2.60   0.01
4/10        4 -802.07 1612.30  2.67   0.01
3/5         4 -802.07 1612.31  2.68   0.01
2/5/10      5 -801.03 1612.32  2.69   0.01
5/7         4 -802.10 1612.36  2.73   0.01
3/4/5/8     6 -800.01 1612.37  2.74   0.01
5/8         4 -802.12 1612.40  2.77   0.01
3/4/9       5 -801.13 1612.51  2.88   0.01
2/3/4/5     6 -800.08 1612.52  2.88   0.01
2/5/6       5 -801.14 1612.54  2.90   0.01
3/4/5/7     6 -800.10 1612.55  2.92   0.01
4/9/10      5 -801.16 1612.58  2.95   0.01
1/5/6       5 -801.18 1612.61  2.98   0.01
5/6/10      5 -801.18 1612.61  2.98   0.01
4/5/8/9     6 -800.14 1612.64  3.01   0.01
2/5/9       5 -801.24 1612.72  3.09   0.01
4/5/6/9     6 -800.20 1612.75  3.12   0.01
1/4/5/6     6 -800.20 1612.76  3.12   0.01
4/5/7/9     6 -800.20 1612.76  3.13   0.01
1/4/9       5 -801.26 1612.76  3.13   0.01
2/4/5/9     6 -800.20 1612.76  3.13   0.01
4/7         4 -802.32 1612.80  3.17   0.01
1/2/5       5 -801.30 1612.85  3.22   0.01
4/7/9       5 -801.30 1612.86  3.23   0.01
1/4/5/10    6 -800.26 1612.87  3.24   0.01
1/4         4 -802.35 1612.88  3.24   0.01
1/3/4       5 -801.34 1612.92  3.29   0.01
5/9/10      5 -801.34 1612.94  3.30   0.01
4/5/6/10    6 -800.30 1612.97  3.33   0.01
2/5/7       5 -801.36 1612.98  3.35   0.01
2/3/5       5 -801.37 1613.00  3.37   0.01
1/4/5/8     6 -800.33 1613.02  3.39   0.01
2/4/5/10    6 -800.35 1613.05  3.42   0.01
4/6/9       5 -801.40 1613.06  3.43   0.01
4/8         4 -802.45 1613.06  3.43   0.01
2/5/8       5 -801.41 1613.08  3.45   0.01
3/5/6       5 -801.42 1613.09  3.46   0.01
4/8/9       5 -801.43 1613.10  3.47   0.01
5/6/9       5 -801.43 1613.11  3.47   0.01
1/3/4/5/6   7 -799.32 1613.12  3.49   0.01
2/4         4 -802.49 1613.15  3.52   0.01
3/4/7       5 -801.46 1613.18  3.54   0.01
4/5/7/10    6 -800.41 1613.18  3.54   0.01
4/5/8/10    6 -800.41 1613.18  3.54   0.01
3/4/8       5 -801.47 1613.19  3.56   0.01
1/2/4/5     6 -800.42 1613.20  3.57   0.01
4/6         4 -802.53 1613.23  3.60   0.01
1/3/4/5/10  7 -799.38 1613.24  3.61   0.01
3/5/10      5 -801.50 1613.25  3.61   0.01
3/4/5/6/10  7 -799.39 1613.27  3.63   0.01
1/4/5/7     6 -800.48 1613.33  3.69   0.01
2/4/9       5 -801.54 1613.33  3.70   0.01
5/8/10      5 -801.54 1613.34  3.70   0.01
3/4/5/9/10  7 -799.43 1613.35  3.72   0.01
5/6/8       5 -801.58 1613.41  3.77   0.01
1/5/10      5 -801.58 1613.41  3.77   0.01
5/6/7       5 -801.58 1613.42  3.79   0.01
4/5/6/8     6 -800.54 1613.44  3.81   0.01
3/4/9/10    6 -800.54 1613.44  3.81   0.01
2/4/5/6     6 -800.56 1613.47  3.84   0.01
5/7/10      5 -801.61 1613.47  3.84   0.01
2/3/4       5 -801.61 1613.47  3.84   0.01
3/4/6       5 -801.62 1613.50  3.87   0.01
1/5/9       5 -801.62 1613.50  3.87   0.01
1/3/4/5/9   7 -799.52 1613.51  3.88   0.01
4/5/6/7     6 -800.59 1613.54  3.91   0.01
1/3/4/5/8   7 -799.54 1613.56  3.93   0.01
2/3/4/5/10  7 -799.56 1613.60  3.97   0.01
2/4/5/7     6 -800.63 1613.63  3.99   0.01

Term codes: 
 x1 x10 x11 x12 x13  x3  x6  x7  x8  x9 
  1   2   3   4   5   6   7   8   9  10 

Model-averaged coefficients: 
             Estimate Std. Error Adjusted SE z value Pr(>|z|)  
(Intercept) 21.726148  10.349691   10.385920   2.092   0.0364 *
x12         -0.060303   0.033535    0.033677   1.791   0.0734 .
x13          0.049389   0.025342    0.025455   1.940   0.0523 .
x11          0.375891   0.353959    0.355619   1.057   0.2905  
x8          -0.009888   0.010048    0.010097   0.979   0.3274  
x9           0.066302   0.073081    0.073449   0.903   0.3667  
x10         -0.134885   0.224293    0.225290   0.599   0.5494  
x1           0.391017   0.548170    0.550904   0.710   0.4778  
x3           1.258309   2.114061    2.123291   0.593   0.5534  
x7          -0.053482   0.218217    0.219239   0.244   0.8073  
x6           0.167158   0.959226    0.963771   0.173   0.8623  

Full model-averaged coefficients (with shrinkage): 
             Estimate Std. Error Adjusted SE z value Pr(>|z|)  
(Intercept) 21.726148  10.349691   10.385920   2.092   0.0364 *
x12         -0.042489   0.039361    0.039446   1.077   0.2814  
x13          0.038541   0.030319    0.030393   1.268   0.2048  
x11          0.103503   0.250382    0.251029   0.412   0.6801  
x8          -0.002376   0.006489    0.006507   0.365   0.7150  
x9           0.015042   0.044529    0.044666   0.337   0.7363  
x10         -0.022982   0.105561    0.105922   0.217   0.8282  
x1           0.069971   0.276107    0.277079   0.253   0.8006  
x3           0.231713   1.029987    1.033478   0.224   0.8226  
x7          -0.006275   0.076700    0.077041   0.081   0.9351  
x6           0.019317   0.330433    0.331957   0.058   0.9536  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Relative variable importance: 
                     x13  x12  x11  x8   x9   x3   x1   x10  x7   x6  
Importance:          0.78 0.70 0.28 0.24 0.23 0.18 0.18 0.17 0.12 0.12
N containing models:   70   65   28   24   24   21   20   19   14   14

看起来不错!如果我运行summary(modav)[3],我会得到:

$avg.model
                Estimate  Std. Error Adjusted SE      Lower CI     Upper CI
(Intercept) 21.726148149 10.34969082 10.38592000  1.3701190119 42.082177286
x12         -0.060303493  0.03353455  0.03367710 -0.1263093944  0.005702409
x13          0.049389493  0.02534158  0.02545464 -0.0005006826  0.099279669
x11          0.375891126  0.35395924  0.35561886 -0.3211090361  1.072891288
x8          -0.009887984  0.01004840  0.01009655 -0.0296768586  0.009900890
x9           0.066302224  0.07308107  0.07344873 -0.0776546368  0.210259085
x10         -0.134884710  0.22429287  0.22529036 -0.5764456948  0.306676276
x1           0.391016656  0.54817001  0.55090396 -0.6887352705  1.470768583
x3           1.258309004  2.11406084  2.12329116 -2.9032652035  5.419883211
x7          -0.053481971  0.21821694  0.21923860 -0.4831817329  0.376217790
x6           0.167158391  0.95922641  0.96377088 -1.7217978265  2.056114609

这正是我想要的。如果我运行summary(modav)[4]而不是获得类似的表格,我只会得到参数估计值。

> summary(modav)[4]
$coef.shrinkage
 (Intercept)          x12          x13          x11           x8           x9          x10           x1           x3           x7           x6 
21.726148149 -0.042488773  0.038540747  0.103502707 -0.002375575  0.015042454 -0.022981800  0.069970584  0.231713307 -0.006274563  0.019317072 

我真正想要的是:

Full model-averaged coefficients (with shrinkage): 
                 Estimate Std. Error Adjusted SE z value Pr(>|z|)  
    (Intercept) 21.726148  10.349691   10.385920   2.092   0.0364 *
    x12         -0.042489   0.039361    0.039446   1.077   0.2814  
    x13          0.038541   0.030319    0.030393   1.268   0.2048  
    x11          0.103503   0.250382    0.251029   0.412   0.6801  
    x8          -0.002376   0.006489    0.006507   0.365   0.7150  
    x9           0.015042   0.044529    0.044666   0.337   0.7363  
    x10         -0.022982   0.105561    0.105922   0.217   0.8282  
    x1           0.069971   0.276107    0.277079   0.253   0.8006  
    x3           0.231713   1.029987    1.033478   0.224   0.8226  
    x7          -0.006275   0.076700    0.077041   0.081   0.9351  
    x6           0.019317   0.330433    0.331957   0.058   0.9536  

任何人都有关于如何获得这个的想法?我不能每次都复制和粘贴,我有60个因变量来运行它。我真的需要每个因变量的每个估计的标准误差。

1 个答案:

答案 0 :(得分:1)

coefTable(modav, full=TRUE) # full=FALSE for "subset" coefficients