Using the 'baltimore' housing data from SpData, I want to model the presence of a patio as the response variable, with house price as the explanatory variable. I also want to include weights in my model by housing area.
My code:
library(spData)
library(nlme)
library(dplyr)
library(MASS)
baltimore<-spData::baltimore
baltimore$logpr = log(baltimore$PRICE)
#alright, i want this to be weighted by sqft
w=baltimore$SQFT/100
w
model1 <- glmmPQL(PATIO ~ PRICE , random = ~1|CITCOU, data = baltimore,family=binomial,correlation = corExp(form = ~X + Y, nugget = T),weights = w)
This basically gives me a different error message for each weighting variable I choose. The use of weights here seem to be the only problem here. The weights vector length is the same as the data in the model, so I don't really understan why this isn't working. Any insight appreciated.
If you make the weights sum to 1, the model converges.
w <- w/sum(w)
model1 <- glmmPQL(PATIO ~ PRICE ,
random = ~1|CITCOU,
data = baltimore,
family=binomial,
correlation = corExp(form = ~X + Y, nugget = T),
weights = w)
summary(model1)
# Linear mixed-effects model fit by maximum likelihood
# Data: baltimore
# AIC BIC logLik
# NA NA NA
#
# Random effects:
# Formula: ~1 | CITCOU
# (Intercept) Residual
# StdDev: 0.001372962 0.06760035
#
# Correlation Structure: Exponential spatial correlation
# Formula: ~X + Y | CITCOU
# Parameter estimate(s):
# range nugget
# 0.03104283 0.11152655
# Variance function:
# Structure: fixed weights
# Formula: ~invwt
# Fixed effects: PATIO ~ PRICE
# Value Std.Error DF t-value p-value
# (Intercept) -4.343533 0.5705149 208 -7.613355 0
# PRICE 0.053687 0.0092687 208 5.792323 0
# Correlation:
# (Intr)
# PRICE -0.937
#
# Standardized Within-Group Residuals:
# Min Q1 Med Q3 Max
# -2.8915877 -0.3851644 -0.2667641 -0.1707177 5.9131663
#
# Number of Observations: 211
# Number of Groups: 2