我正在使用COVID-19主题进行单项评估，并试图对COVID 19大流行进行泊松回归分析。
[我的预测变量是计数的确诊病例数，我在泊松回归中的预测变量是outbreak_days，apple_drivingmobility，国家/地区。
然后，模型附带了巨大的AIC，也有偏差。
那是因为我的犯罪案件是不遵循泊松分布的累积量？相反，我应该每天使用新案例吗？或这完全不适合使用泊松。
glm(formula = confirmed_deaths ~ outbreak_days + Apple_DrivingMobility +
country, family = poisson(link = log), data = df_cleaned_driving_level)
Deviance Residuals:
Min 1Q Median 3Q Max
-131.981 -9.773 -0.863 3.798 87.411
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 1.726e+00 1.795e-02 96.183 <2e-16 ***
outbreak_days 6.706e-02 3.687e-05 1818.954 <2e-16 ***
Apple_DrivingMobility -1.847e-02 4.573e-05 -403.834 <2e-16 ***
countryBelgium 4.404e+00 1.791e-02 245.852 <2e-16 ***
countryBrazil 4.604e+00 1.789e-02 257.378 <2e-16 ***
countryCanada 3.755e+00 1.806e-02 207.950 <2e-16 ***
countryCzechia 1.746e+00 2.048e-02 85.283 <2e-16 ***
countryDenmark 2.431e+00 1.934e-02 125.720 <2e-16 ***
countryEstonia 4.712e-01 2.768e-02 17.020 <2e-16 ***
countryFinland 1.693e+00 2.109e-02 80.292 <2e-16 ***
countryFrance 4.922e+00 1.788e-02 275.355 <2e-16 ***
countryGermany 4.186e+00 1.796e-02 232.992 <2e-16 ***
countryIreland 2.447e+00 1.839e-02 133.051 <2e-16 ***
countryItaly 4.871e+00 1.791e-02 272.003 <2e-16 ***
countryJapan 8.318e-01 2.285e-02 36.399 <2e-16 ***
countryLuxembourg 2.099e-01 2.338e-02 8.979 <2e-16 ***
countryMexico 3.551e+00 1.806e-02 196.676 <2e-16 ***
countryNetherlands 4.098e+00 1.797e-02 228.088 <2e-16 ***
countryNewZealand -1.148e+00 4.103e-02 -27.984 <2e-16 ***
countryNorway 1.440e+00 2.145e-02 67.110 <2e-16 ***
countryPhilippines 1.448e+00 1.893e-02 76.457 <2e-16 ***
countrySingapore -2.988e+00 7.258e-02 -41.170 <2e-16 ***
countrySlovakia -8.289e-01 3.875e-02 -21.393 <2e-16 ***
countrySpain 4.975e+00 1.790e-02 277.988 <2e-16 ***
countrySweden 4.060e+00 1.827e-02 222.273 <2e-16 ***
countrySwitzerland 3.208e+00 1.829e-02 175.383 <2e-16 ***
countryTaiwan* -1.535e+00 5.629e-02 -27.272 <2e-16 ***
countryUnitedKingdom 5.153e+00 1.785e-02 288.619 <2e-16 ***
countryUnitedStates 6.537e+00 1.783e-02 366.664 <2e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Dispersion parameter for poisson family taken to be 1)
Null deviance: 20309689 on 1879 degrees of freedom
Residual deviance: 765387 on 1851 degrees of freedom
AIC: 777730
Number of Fisher Scoring iterations: 6
[![enter image description here][1]][1]
AIC取决于对数似然性：AIC = 2k-2 ln（L），其中L是对数似然性。偏差-2 * L，k是模型中参数的数量。