error when adding lm to datetime data in ggplot2-CodePudding

When Trying to plot the linear regression of the data the output returns a horizontal line with a warning that the fit is rank-deficient. Can someone please explain how to display a proper linear regression?

Sample Data

ToPlot <- structure(list(Time = structure(c(1643126476.049, 1643126476.099, 
1643126476.149, 1643126476.199, 1643126476.249, 1643126476.299, 
1643126476.349, 1643126476.399, 1643126476.449, 1643126476.499, 
1643126476.549, 1643126476.599, 1643126476.649, 1643126476.699, 
1643126476.749, 1643126476.799, 1643126476.849, 1643126476.899, 
1643126476.949, 1643126476.999), class = c("POSIXct", "POSIXt"
), tzone = ""), LIN = c(112.05611159876, 112.66549348, 113.37316276144, 
113.86459976244, 114.59192652392, 115.29959580536, 115.88932020656, 
116.61664696804, 117.20637136924, 117.93369813072, 118.52342253192, 
119.13280441316, 119.8404736946, 120.54814297604, 121.15752485728, 
121.86519413872, 122.47457601996, 123.1822453014, 123.79162718264, 
124.49929646408)), row.names = 314:333, class = "data.frame")

Code:

library("ggplot2")
library("scales")

ggplot(aes(x = Time, y = LIN), data = ToPlot)   
  geom_point()   
  geom_line()   
  scale_x_datetime(labels = date_format("%H:%m:%S")) 
  geom_smooth(method='lm')

CodePudding user response：

Typically the package picks the orientation (x or y) automatically. However, when it becomes uncertain, you see that message. You don't need to specify the formula, you can just specify the orientation.

library("ggplot2")
library("scales")

ggplot(aes(x = Time, y = LIN), data = ToPlot)   
  geom_point()   
  geom_line()   
  scale_x_datetime(labels = date_format("%H:%m:%S")) 
  geom_smooth(method='lm', orientation = "y")

CodePudding user response：

This isn't a ggplot problem, but an lm problem:

lm(LIN ~ Time, data = ToPlot)
#> 
#> Call:
#> lm(formula = LIN ~ Time, data = ToPlot)
#> 
#> Coefficients:
#> (Intercept)         Time  
#>      118.2           NA

I think the issue here isn't the POSIXct format, but just that it seems to choke on the large numbers for some reason (you get the same result if you convert Time to numeric).

One way to get round this is simply to subtract the large constant number of seconds in the POSIXct within the formula argument:

ToPlot <- structure(list(Time = structure(c(1643126476.049, 1643126476.099, 
1643126476.149, 1643126476.199, 1643126476.249, 1643126476.299, 
1643126476.349, 1643126476.399, 1643126476.449, 1643126476.499, 
1643126476.549, 1643126476.599, 1643126476.649, 1643126476.699, 
1643126476.749, 1643126476.799, 1643126476.849, 1643126476.899, 
1643126476.949, 1643126476.999), class = c("POSIXct", "POSIXt"
), tzone = ""), LIN = c(112.05611159876, 112.66549348, 113.37316276144, 
113.86459976244, 114.59192652392, 115.29959580536, 115.88932020656, 
116.61664696804, 117.20637136924, 117.93369813072, 118.52342253192, 
119.13280441316, 119.8404736946, 120.54814297604, 121.15752485728, 
121.86519413872, 122.47457601996, 123.1822453014, 123.79162718264, 
124.49929646408)), row.names = 314:333, class = "data.frame")

library("ggplot2")
library("scales")

datemin <- floor(as.numeric(min(ToPlot$Time)))

ggplot(aes(x = Time, y = LIN), data = ToPlot)   
  geom_point()   
  geom_line()   
  scale_x_datetime(labels = date_format("%H:%m:%S")) 
  geom_smooth(formula = y ~ I(x - datemin), method='lm')

^{Created on 2022-01-27 by the reprex package (v2.0.1)}