library(readxl)
library(lmtest)
## Loading required package: zoo
##
## Attaching package: 'zoo'
## The following objects are masked from 'package:base':
##
## as.Date, as.Date.numeric
library(sandwich)
library(car)
## Loading required package: carData
library(tidyverse)
## ── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
## ✔ dplyr 1.2.1 ✔ readr 2.2.0
## ✔ forcats 1.0.1 ✔ stringr 1.6.0
## ✔ ggplot2 4.0.3 ✔ tibble 3.3.1
## ✔ lubridate 1.9.5 ✔ tidyr 1.3.2
## ✔ purrr 1.2.2
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag() masks stats::lag()
## ✖ dplyr::recode() masks car::recode()
## ✖ purrr::some() masks car::some()
## ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errors
district<-read_excel("district.xls")
district <- district[district$COMMTYPE != "Charters", ]
district <- district[!is.na(district$DAGC4X21R) & district$DAGC4X21R >= 0, ]
district <- district[!is.na(district$DPFVTOTK) & district$DPFVTOTK > 0, ]
#removing charters, districts with missing/hidden graduation rates, and those with no property value
district <- district |>
mutate(grad = DAGC4X21R, # dependent variable: 4-year graduation rate
tax_value = DPFVTOTK, # independent variable: taxable value per pupil
log_tax = log(DPFVTOTK), # logged version used in the models
econ = DPETECOP, # % economically disadvantaged
enroll = DPETALLC, # total students
log_enroll = log(DPETALLC),
# urban = 1 if the district is one of the four city or suburb types, 0 if not
urban = ifelse(COMMTYPE == "Major Urban", 1, 0) +
ifelse(COMMTYPE == "Major Suburban", 1, 0) +
ifelse(COMMTYPE == "Other Central City", 1, 0) +
ifelse(COMMTYPE == "Other Central City Suburban", 1, 0))
model1 <- lm(grad ~ log_tax, data = district)
model2 <- lm(grad ~ log_tax + econ + log_enroll + urban, data = district)
summary(model1)
##
## Call:
## lm(formula = grad ~ log_tax, data = district)
##
## Residuals:
## Min 1Q Median 3Q Max
## -36.105 -2.034 1.216 3.823 5.145
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 90.3139 2.7675 32.634 <2e-16 ***
## log_tax 0.4204 0.2105 1.997 0.0461 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 4.739 on 958 degrees of freedom
## Multiple R-squared: 0.004144, Adjusted R-squared: 0.003105
## F-statistic: 3.987 on 1 and 958 DF, p-value: 0.04614
summary(model2)
##
## Call:
## lm(formula = grad ~ log_tax + econ + log_enroll + urban, data = district)
##
## Residuals:
## Min 1Q Median 3Q Max
## -36.905 -1.430 1.050 2.655 7.310
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 114.83064 3.00150 38.258 < 2e-16 ***
## log_tax -0.52643 0.19867 -2.650 0.00819 **
## econ -0.08662 0.00739 -11.722 < 2e-16 ***
## log_enroll -0.94947 0.13121 -7.236 9.46e-13 ***
## urban -0.24827 0.43352 -0.573 0.56699
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 4.238 on 955 degrees of freedom
## Multiple R-squared: 0.2062, Adjusted R-squared: 0.2029
## F-statistic: 62.01 on 4 and 955 DF, p-value: < 2.2e-16
#regression models (table 2)
model_raw <- lm(grad ~ tax_value, data = district)
plot(model_raw, which = 1)

raintest(model_raw)
##
## Rainbow test
##
## data: model_raw
## Rain = 0.7412, df1 = 480, df2 = 478, p-value = 0.9995
# After the log transformation
plot(model1, which = 1)

raintest(model1)
##
## Rainbow test
##
## data: model1
## Rain = 0.74261, df1 = 480, df2 = 478, p-value = 0.9994
raintest(model2)
##
## Rainbow test
##
## data: model2
## Rain = 0.78951, df1 = 480, df2 = 475, p-value = 0.995
# Linearity: residual & rainbow test then log
plot(model2, which = 3)

bptest(model1)
##
## studentized Breusch-Pagan test
##
## data: model1
## BP = 1.2239, df = 1, p-value = 0.2686
bptest(model2)
##
## studentized Breusch-Pagan test
##
## data: model2
## BP = 29.678, df = 4, p-value = 5.693e-06
#Homoscedasticity: plot and BP test
durbinWatsonTest(model1)
## lag Autocorrelation D-W Statistic p-value
## 1 0.06211333 1.873778 0.05
## Alternative hypothesis: rho != 0
durbinWatsonTest(model2)
## lag Autocorrelation D-W Statistic p-value
## 1 0.017684 1.963723 0.542
## Alternative hypothesis: rho != 0
# Independence of errors: Durbin-Watson test
plot(model1, which = 2)

shapiro.test(model1$residuals)
##
## Shapiro-Wilk normality test
##
## data: model1$residuals
## W = 0.82895, p-value < 2.2e-16
plot(model2, which = 2)

shapiro.test(model2$residuals)
##
## Shapiro-Wilk normality test
##
## data: model2$residuals
## W = 0.83589, p-value < 2.2e-16
# Normality of errors: Q-Q plot and Shapiro-Wilk test
vif(model2)
## log_tax econ log_enroll urban
## 1.113669 1.097808 2.108636 2.088264
# Multicollinearity
model2_log <- lm(log(grad) ~ log_tax + econ + log_enroll + urban, data = district)
bptest(model2_log) # still p < .001, so no bueno
##
## studentized Breusch-Pagan test
##
## data: model2_log
## BP = 24.634, df = 4, p-value = 5.961e-05
#fixing the unequal spread
0.4204 * log(2)
## [1] 0.2913991
-0.5264 * log(2)
## [1] -0.3648727
# doubling property wealth (coefficient x log of 2)
-0.0866 * 10 # about -0.9 points
## [1] -0.866
# adding 10 more points of economically disadvantaged students