Open rsbivand opened 1 year ago
Hi, I have this issue when using predict
on a Durbin model with only some lagged variables. Assuming the problem isn't easy to fix, would there be any other way to generate forecasts with a model like this? Thanks.
Could you provide a reproducible example that matches your case?
So after looking at it more carefully, I believe my error was actually being caused by a mismatch of structure between my original dataset and the predicting dataset. The message I had was:
"Error in predict.Sarlm(mod, dfp, listw = weights) : unknown mismatch. please report this bug In addition: Warning message: In colnames(Xs.not.lagged) != colnames(Xo) : longer object length is not a multiple of shorter object length"
I assumed the first error (which in the code you link to this issue) was the main problem but fixing the second warning appears to have made everything work. An example of what I'm doing:
library(tidyverse)
library(sf)
library(spdep)
library(spatialreg)
# Dataset
df <- st_read(system.file("shapes/boston_tracts.shp", package = "spData"))
# Spatial weights
centroids <- st_centroid(df)
neigh <- dnearneigh(centroids, d1 = 0, d2 = 30)
weights <- nb2listw(neigh, glist = lapply(nbdists(neigh, centroids), \(x) 1/(x^2)))
# Model
mod <- lagsarlm(log(MEDV) ~ CRIM + ZN + INDUS + CHAS + I(NOX^2) + I(RM^2) +
AGE + log(DIS) + log(RAD) + TAX + PTRATIO + B + log(LSTAT),
data = df,
Durbin = ~ CRIM,
listw = weights)
summary(mod)
# Predict
dfp <- df %>%
mutate(TAX = 0.5 * TAX)
predict(mod, dfp, listw = weights)
I think this works fine. But just to be clear, predict
is aware that only CRIM
is a spatially lagged variable in the model, as specified in lagsarlm
, right? I think I may have misunderstood this issue.
Yes, predict
only uses lag.CRIM
, not other lagged X variables.
Handling of formula Durbin objects does not work, get to line 233 in R/predict.sarlm.R.