使用 by() 从模型中提取预测值

Question

I'm fitting an exponential model to population data for 208 springsheds to back-calculate population for years 1975-2015 in 5-year intervals, ie seq(1975,2015,5) . Here's the first 5 springs in my dataset and the code I'm using to fit the model and plot it (I want to have the figures):

springsheds <- 
structure(list(spring = c("alexander", "alexander", "alexander", "alexander", 
"blue hole", "blue hole", "blue hole", "blue hole", "cedar head", "cedar head", 
"cedar head", "cedar head", "charles", "charles", "charles", "charles", 
"columbia", "columbia", "columbia", "columbia"), year = c(2000L, 2005L, 2010L, 
2015L, 2000L, 2005L, 2010L, 2015L, 2000L, 2005L, 2010L, 2015L, 2000L, 2005L, 
2010L, 2015L, 2000L, 2005L, 2010L, 2015L), pop = c(527L, 620L, 732L, 867L, 
3071L, 3356L, 3669L, 4007L, 3038L, 3320L, 3630L, 3965L, 1311L, 1446L, 1592L, 
1747L, 7550L, 8130L, 8706L, 9332L)), .Names = c("spring", "year", "pop"), 
class = "data.frame", row.names = c(NA, -20L))

models.spsh <- by(springsheds, springsheds$spring, function(x) {
        fm <- lm(log(pop) ~ year, data = x)
        timevalues <- seq(1970, 2020, 10)
        predict <- exp(predict(fm,list(year=timevalues)))
        plot(pop ~ year, x, main = spring[1], xlim = c(1970, 2020), ylim=c(0,15000))
        lines(timevalues, predict,lwd=1, col = "blue", xlab = "Year", ylab = "Population")
})

Can I also use by() to extract the predicted values for each spring? My current workaround is to create an object for each spring separately and iteratively add the predicted values to an object:

fm <- lm(log(pop) ~ year, data = alex)
timevalues <- seq(1975,2015,5)
alex <- exp(predict(fm,list(year=timevalues)))
old<-cbind(timevalues,alex)
fm <- lm(log(pop) ~ year, data = blue)
blue <- exp(predict(fm,list(year=timevalues)))
old<-cbind(old,blue)

This seems really inefficient and I'm assuming there's a more elegant way of doing this, is there a way I can just add to my initial code to also extract the predicted population values?

Answer 1

You can split the data and then use lapply for each desired output:

#Split the data grouped by spring
sdata <- split(springsheds, springsheds$spring)

#Fit the models
fit.spsh <- lapply(sdata, function(x) {
  lm(log(pop) ~ year, data = x)
})

#Get the predicted values
timevalues <- seq(1970, 2020, 10)
predictList <- lapply(fit.spsh, function(m) exp(predict(m,list(year=timevalues))))

#Generate plots
lapply(names(sdata), function(n) {
  plot(pop ~ year,sdata[[n]] , main = n, xlim = c(1970, 2020), ylim=c(0,15000))
  lines(timevalues, predictList[[n]],lwd=1, col = "blue", xlab = "Year", ylab = "Population")

})

 #Combine the predict values
 do.call(cbind,predictList)
 #alexander blue hole cedar head   charles  columbia  
 #1  194.3679  1803.470   1783.068  738.9545  4955.633
 #2  270.8778  2153.663   2129.682  894.9253  5705.076
 #3  377.5048  2571.856   2543.676 1083.8167  6567.857
 #4  526.1037  3071.253   3038.146 1312.5774  7561.118
 #5  733.1965  3667.621   3628.738 1589.6225  8704.590
 #6 1021.8081  4379.790   4334.137 1925.1434 10020.989