Added modeling file and plots

danielkberry · Nov 17, 2016 · ad2aa32 · ad2aa32
1 parent 4c6acc5
commit ad2aa32
Show file tree

Hide file tree

Showing 59 changed files with 134 additions and 1 deletion.
diff --git a/ETL.r b/ETL.r
@@ -170,6 +170,13 @@ socioeconomic <- read.csv('Census_Data_-_Selected_socioeconomic_indicators_in_Ch
 
 race <- read.csv('race.csv', stringsAsFactors = FALSE)
 
+## NHW: Non-hispanic white
+## NHB: non-hispanic black
+## NHAM: american indian/alaskan native, non hispanic
+## NHAS: asian, not hispanic
+## NHOTHER: other single race, not hispanic
+
+
 block_data$Neighborhood <- as.character(block_data$Neighborhood)
 
 ## Set up table to match subneighborhoods

diff --git a/Plot of Assault..Homicide..png b/Plot of Assault..Homicide..png
diff --git a/Plot of Below.Poverty.Level.png b/Plot of Below.Poverty.Level.png
diff --git a/Plot of Birth.Rate.png b/Plot of Birth.Rate.png
diff --git a/Plot of Breast.cancer.in.females.png b/Plot of Breast.cancer.in.females.png
diff --git a/Plot of CTA_counts.png b/Plot of CTA_counts.png
diff --git a/Plot of Cancer..All.Sites..png b/Plot of Cancer..All.Sites..png
diff --git a/Plot of Childhood.Blood.Lead.Level.Screening.png b/Plot of Childhood.Blood.Lead.Level.Screening.png
diff --git a/Plot of Childhood.Lead.Poisoning.png b/Plot of Childhood.Lead.Poisoning.png
diff --git a/Plot of Colorectal.Cancer.png b/Plot of Colorectal.Cancer.png
diff --git a/Plot of Community.Area.Number.png b/Plot of Community.Area.Number.png
diff --git a/Plot of Community.Area.x.png b/Plot of Community.Area.x.png
diff --git a/Plot of Crowded.Housing.png b/Plot of Crowded.Housing.png
diff --git a/Plot of Dependency.png b/Plot of Dependency.png
diff --git a/Plot of Diabetes.related.png b/Plot of Diabetes.related.png
diff --git a/Plot of Firearm.related.png b/Plot of Firearm.related.png
diff --git a/Plot of General.Fertility.Rate.png b/Plot of General.Fertility.Rate.png
diff --git a/Plot of Gonorrhea.in.Females.png b/Plot of Gonorrhea.in.Females.png
diff --git a/Plot of Gonorrhea.in.Males.png b/Plot of Gonorrhea.in.Males.png
diff --git a/Plot of HARDSHIP.INDEX.png b/Plot of HARDSHIP.INDEX.png
diff --git a/Plot of HISP.png b/Plot of HISP.png
diff --git a/Plot of HISP_p.png b/Plot of HISP_p.png
diff --git a/Plot of Infant.Mortality.Rate.png b/Plot of Infant.Mortality.Rate.png
diff --git a/Plot of Low.Birth.Weight.png b/Plot of Low.Birth.Weight.png
diff --git a/Plot of Lung.Cancer.png b/Plot of Lung.Cancer.png
diff --git a/Plot of Multiple.Race...png b/Plot of Multiple.Race...png
diff --git a/Plot of Multiple.Race.._p.png b/Plot of Multiple.Race.._p.png
diff --git a/Plot of NHAM.png b/Plot of NHAM.png
diff --git a/Plot of NHAM_p.png b/Plot of NHAM_p.png
diff --git a/Plot of NHAS.png b/Plot of NHAS.png
diff --git a/Plot of NHAS_p.png b/Plot of NHAS_p.png
diff --git a/Plot of NHB.png b/Plot of NHB.png
diff --git a/Plot of NHB_p.png b/Plot of NHB_p.png
diff --git a/Plot of NHOTHER.png b/Plot of NHOTHER.png
diff --git a/Plot of NHOTHER_p.png b/Plot of NHOTHER_p.png
diff --git a/Plot of NHW.png b/Plot of NHW.png
diff --git a/Plot of NHW_p.png b/Plot of NHW_p.png
diff --git a/Plot of No.High.School.Diploma.png b/Plot of No.High.School.Diploma.png
diff --git a/Plot of PER.CAPITA.INCOME.png b/Plot of PER.CAPITA.INCOME.png
diff --git a/Plot of PERCENT.AGED.16..UNEMPLOYED.png b/Plot of PERCENT.AGED.16..UNEMPLOYED.png
diff --git a/Plot of PERCENT.AGED.25..WITHOUT.HIGH.SCHOOL.DIPLOMA.png b/Plot of PERCENT.AGED.25..WITHOUT.HIGH.SCHOOL.DIPLOMA.png
diff --git a/Plot of PERCENT.AGED.UNDER.18.OR.OVER.64.png b/Plot of PERCENT.AGED.UNDER.18.OR.OVER.64.png
diff --git a/Plot of PERCENT.HOUSEHOLDS.BELOW.POVERTY.png b/Plot of PERCENT.HOUSEHOLDS.BELOW.POVERTY.png
diff --git a/Plot of PERCENT.OF.HOUSING.CROWDED.png b/Plot of PERCENT.OF.HOUSING.CROWDED.png
diff --git a/Plot of Prenatal.Care.Beginning.in.First.Trimester.png b/Plot of Prenatal.Care.Beginning.in.First.Trimester.png
diff --git a/Plot of Preterm.Births.png b/Plot of Preterm.Births.png
diff --git a/Plot of Prostate.Cancer.in.Males.png b/Plot of Prostate.Cancer.in.Males.png
diff --git a/Plot of Stroke..Cerebrovascular.Disease..png b/Plot of Stroke..Cerebrovascular.Disease..png
diff --git a/Plot of TOTAL.POPULATION.png b/Plot of TOTAL.POPULATION.png
diff --git a/Plot of TOTAL.png b/Plot of TOTAL.png
diff --git a/Plot of Teen.Birth.Rate.png b/Plot of Teen.Birth.Rate.png
diff --git a/Plot of Tuberculosis.png b/Plot of Tuberculosis.png
diff --git a/Plot of Unemployment.png b/Plot of Unemployment.png
diff --git a/Plot of desert.png b/Plot of desert.png
diff --git a/Plot of nearest_supermarket.png b/Plot of nearest_supermarket.png
diff --git a/Plot of store_counts.png b/Plot of store_counts.png
diff --git a/Plot of vacant_counts.png b/Plot of vacant_counts.png
diff --git a/model.r b/model.r
@@ -0,0 +1,86 @@
+########################
+## Modeling code file ##
+########################
+
+## load packages
+library(lme4)
+
+## load data
+load('all_data')
+all_data$desert <- as.numeric(all_data$desert)
+
+model_data <- subset(all_data, TOTAL.POPULATION > 0)
+
+
+######################
+## Complete pooling ##
+######################
+
+cp <- glm(desert ~ CTA_counts + vacant_counts, family = 'binomial', data = model_data)
+summary(cp)
+################
+## No pooling ##
+################
+
+np <- glm(desert ~ CTA_counts + vacant_counts + Neighborhood, family = 'binomial', data = model_data)
+summary(np)
+
+#####################
+## Partial pooling ##
+#####################
+
+pp <- glmer(desert ~ CTA_counts + vacant_counts + (1 | Neighborhood), data = model_data, family = 'binomial')
+summary(pp)
+
+##################
+## Hierarchical ##
+##################
+
+mlm <- glmer(desert ~ CTA_counts + vacant_counts + Diabetes.related + (1 | Neighborhood),
+             data = model_data,
+             family = 'binomial')
+summary(mlm)
+
+mlm_2 <- glmer(desert ~ CTA_counts + vacant_counts + Diabetes.related + Below.Poverty.Level + (1 | Neighborhood),
+             data = model_data,
+             family = 'binomial')
+summary(mlm_2)
+
+
+## rescale variables
+exclude <- c('Neighborhood', 'TRACT_BLOC','STATEFP10', 'COUNTYFP10', 'TRACTCE10', 'BLOCKCE10', 'GEOID10', 'NAME10', 'Longitude', 'Latitude', 'Community.Area.y', 'nearest_supermarket', 'Community.Area.x', 'store_counts', 'desert')
+potential_covariates <- setdiff(names(all_data), exclude)
+model_data_scale <- model_data
+for (var in potential_covariates) {model_data_scale[var] <- as.numeric(scale(model_data[var]))}
+
+mlm_c <- glmer(desert ~ CTA_counts + vacant_counts + Diabetes.related + (1 | Neighborhood),
+             data = model_data_scale,
+             family = 'binomial')
+summary(mlm_c)
+
+mlm_c_2 <- glmer(desert ~ CTA_counts + vacant_counts + Diabetes.related + Below.Poverty.Level + (1 | Neighborhood),
+             data = model_data_scale,
+             family = 'binomial')
+summary(mlm_c_2)
+
+mlm_c_3 <- glmer(desert ~ CTA_counts + vacant_counts + Diabetes.related + Below.Poverty.Level + NHB_p + PER.CAPITA.INCOME + (1 | Neighborhood),
+             data = model_data_scale,
+             family = 'binomial')
+summary(mlm_c_3)
+
+mlm_c_4 <- glmer(desert ~ CTA_counts + vacant_counts + Diabetes.related + Below.Poverty.Level + NHB_p + PER.CAPITA.INCOME + HISP_p + (1 | Neighborhood),
+             data = model_data_scale,
+             family = 'binomial')
+summary(mlm_c_4)
+
+mlm_c_5 <- glmer(desert ~ CTA_counts + vacant_counts + Diabetes.related + Below.Poverty.Level + NHB_p + PER.CAPITA.INCOME + HISP_p + TOTAL.POPULATION + (1 | Neighborhood),
+             data = model_data_scale,
+             family = 'binomial')
+summary(mlm_c_5)
+
+mlm_c_5 <- glmer(desert ~ CTA_counts + vacant_counts + Diabetes.related + Below.Poverty.Level + NHW_p + NHB_p + HISP_p + PER.CAPITA.INCOME  + TOTAL.POPULATION + (1 | Neighborhood),
+             data = model_data_scale,
+             family = 'binomial')
+summary(mlm_c_5)
+
+cor(model_data[, c('Diabetes.related', 'NHB_p', 'NHW_p', 'HISP_p')])
diff --git a/plots.r b/plots.r
@@ -7,7 +7,7 @@ load('all_data')
 ## Plot location of missing blocks
 library(ggplot2)
 
-pct_missing <- lapply(all_data, function(var) mean(is.na(var)))
+pct_missing <- lapply(all_data, function(var) sum(is.na(var)))
 
 plot_data <- all_data
 plot_data$missing <- is.na(plot_data$Neighborhood)
@@ -57,3 +57,43 @@ t <- project(as.matrix(nbhd_df[,c('long', 'lat')]),'+proj=merc +lat_0=36.6666666
 head(t <- project(as.matrix(nbhd_df[,c('long', 'lat')]),'+proj=merc +lat_0=36.66666666666666+lon_0=-88.33333333333333', inv = TRUE))
 
 head(t <- project(as.matrix(nbhd_df[,c('long', 'lat')]),'+proj=merc +lat_0=36.66666666666666 +lon_0=-88.33333333333333 +k=0.9999749999999999 +x_0=300000 +y_0=0', inv = TRUE))
+
+
+######################
+## Univariate Plots ##
+######################
+
+all_data$desert <- as.numeric(all_data$desert)
+
+
+
+
+
+summary(glm(desert ~ NHB_p, family = 'binomial', data = all_data))
+
+ggplot(all_data, aes(x = NHB_p, y = desert)) +
+    geom_point() + stat_smooth(method = 'glm', method.args = list(family = 'binomial'))
+
+
+summary(glm(desert ~ vacant_counts, family = 'binomial', data = all_data))
+
+ggplot(all_data, aes(x = vacant_counts, y = desert)) +
+    geom_point() + stat_smooth(method = 'glm', method.args = list(family = 'binomial'))
+
+exclude <- c('Neighborhood', 'TRACT_BLOC','STATEFP10', 'COUNTYFP10', 'TRACTCE10', 'BLOCKCE10', 'GEOID10', 'NAME10', 'Longitude', 'Latitude', 'Community.Area.y', 'nearest_supermarket', 'desert', 'Community.Area.x', 'store_counts')
+potential_covariates <- setdiff(names(all_data), exclude)
+
+for (covar in potential_covariates) {
+    print(covar)
+    covar_plot <- ggplot(all_data, aes_string(x = covar, y = 'desert')) +
+        geom_point() + stat_smooth(method = 'glm', method.args = list(family = 'binomial')) +
+        theme_bw()
+    ggsave(paste0('Plot of ',covar, '.png'), covar_plot)
+}
+
+##########
+## MAPS ##
+##########
+
+all_data$desert_logical <- all_data$desert == 1
+ggplot(all_data, aes(x = Longitude, y = Latitude, color = desert_logical)) + geom_point(alpha = .1) + theme_bw()