我想知道是否有一种更简洁的方式,而不仅仅是虚拟编码月(例如,isJan,isFeb ......),以获得更有意义的自变量名称(在拦截下).我的数据集相当大,所以我在这里模拟了一个简单的数据集.
#create simulated data set with sales, and date
sales <- rnorm(1000, mean = 1000, sd = 40)
dates <- seq(from = 14610, to = 15609)
data <- cbind(sales, dates)
#regression with months
model <- lm(sales ~ months(dates))
summary(model)
我想拦截标签显示他们引用的实际月份...目前我的输出看起来像这样:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 999.1934 1.2673 788.432 <2e-16 ***
months(dates).L -4.9537 4.5689 -1.084 0.2785
months(dates).Q -6.4931 4.4211 -1.469 0.1422
months(dates).C -5.5078 4.4180 -1.247 0.2128
months(dates)^4 2.3713 4.4864 0.529 0.5972
months(dates)^5 -1.7749 4.4605 -0.398 0.6908
months(dates)^6 1.5774 4.4555 0.354 0.7234
months(dates)^7 -10.9954 4.4511 -2.470 0.0137 *
months(dates)^8 -0.9627 4.4032 -0.219 0.8270
months(dates)^9 1.8847 4.2996 0.438 0.6612
months(dates)^10 -8.5990 4.1776 -2.058 0.0398 *
months(dates)^11 7.8436 4.1292 1.900 0.0578 .
在此先感谢, - .JT
你遇到的问题是R已经创建了一个有序因子,并且对于有序因子产生的对比与多项式形成对比(.L是线性的,.Q是二次的,.C立方的并且.^n是n阶多项式.将月定义为一个月可能更好.因子,将第一级设置为1月,然后拟合模型.
如果在英语语言环境中,那么我们可以使用month.name或month.abb常量如下
set.seed(42)
dat <- data.frame(sales = rnorm(1000, mean = 1000, sd = 40),
dates = as.Date(seq(from = 14610, to = 15609),
origin = "1970-01-01"))
dat <- transform(dat, month = factor(format(dates, format = "%B"),
levels = month.name))
这给了
> head(dat)
sales dates month
1 1054.8383 2010-01-01 January
2 977.4121 2010-01-02 January
3 1014.5251 2010-01-03 January
4 1025.3145 2010-01-04 January
5 1016.1707 2010-01-05 January
6 995.7550 2010-01-06 January
> with(dat, levels(month))
[1] "January" "February" "March" "April" "May"
[6] "June" "July" "August" "September" "October"
[11] "November" "December"
请注意,级别的顺序是逻辑顺序而不是字母顺序.如果您使用的是非英语语言环境,则输出"%B"将是您当地语言或约定中的月份名称.然后,您需要为levels上面代码中的参数提供正确的级别作为字符向量.
然后可以使用该数据集来拟合模型,并获得更有意义的系数名称
> mod <- lm(sales ~ month, data = dat)
> summary(mod)
Call:
lm(formula = sales ~ month, data = dat)
Residuals:
Min 1Q Median 3Q Max
-140.333 -24.551 0.108 28.102 134.349
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 1001.7034 4.1567 240.983 <2e-16 ***
monthFebruary -8.3618 6.0153 -1.390 0.165
monthMarch -0.5347 5.8785 -0.091 0.928
monthApril -7.5618 5.9273 -1.276 0.202
monthMay -2.2961 5.8785 -0.391 0.696
monthJune 3.5091 5.9273 0.592 0.554
monthJuly -4.9975 5.8785 -0.850 0.395
monthAugust -0.3558 5.8785 -0.061 0.952
monthSeptember 3.7597 5.9970 0.627 0.531
monthOctober -2.5948 6.5724 -0.395 0.693
monthNovember -10.5670 6.6378 -1.592 0.112
monthDecember -6.9064 6.5724 -1.051 0.294
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 40.09 on 988 degrees of freedom
Multiple R-squared: 0.01173, Adjusted R-squared: 0.0007317
F-statistic: 1.066 on 11 and 988 DF, p-value: 0.3854
在上文中,请注意1月是第一个水平,因此其平均值是(Intercept)估计值,其他估计值与1月平均值的偏差.模型的另一个参数化是抑制截距:
> mod2 <- lm(sales ~ month - 1, data = dat)
> summary(mod2)
Call:
lm(formula = sales ~ month - 1, data = dat)
Residuals:
Min 1Q Median 3Q Max
-140.333 -24.551 0.108 28.102 134.349
Coefficients:
Estimate Std. Error t value Pr(>|t|)
monthJanuary 1001.703 4.157 241.0 <2e-16 ***
monthFebruary 993.342 4.348 228.5 <2e-16 ***
monthMarch 1001.169 4.157 240.9 <2e-16 ***
monthApril 994.142 4.225 235.3 <2e-16 ***
monthMay 999.407 4.157 240.4 <2e-16 ***
monthJune 1005.213 4.225 237.9 <2e-16 ***
monthJuly 996.706 4.157 239.8 <2e-16 ***
monthAugust 1001.348 4.157 240.9 <2e-16 ***
monthSeptember 1005.463 4.323 232.6 <2e-16 ***
monthOctober 999.109 5.091 196.3 <2e-16 ***
monthNovember 991.136 5.175 191.5 <2e-16 ***
monthDecember 994.797 5.091 195.4 <2e-16 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Residual standard error: 40.09 on 988 degrees of freedom
Multiple R-squared: 0.9984, Adjusted R-squared: 0.9984
F-statistic: 5.175e+04 on 12 and 988 DF, p-value: < 2.2e-16
现在,估计数是月度均值,而t检验是个体月均值为零(0)的假设.