[R-sig-ME] Getting the error message: Error: (maxstephalfit) PIRLS step-halvings failed to reduce deviance in pwrssUpdate when trying to fit glmer to my data

[Nimrod Rubinstein]

Hi,

I have these data:

my.df = structure(list(response = c(0.516074952, 0.580012258, 0.5488396,
0.514189282, 0.541119544, 0.534505123, 0.51646652, 0.509050733,
0.518528396, 0.518612368, 0.633614104, 0.571088979, 0.707287635,
0.551960442, 0.628561587, 0.636816662, 0.84273424, 0.791398785,
0.564327876, 0.572247664, 0.542536414, 0.817166116, 0.598312173,
0.549507424, 0.648524073, 0.548890252, 0.525518275, 0.55938954,
0.632503362, 0.525381357, 0.52893369), distance = c(558648, 384605, 357615,
313157, 268367, 201775, 166657, 145176, 143358, 92778, 80661, 64350, 33872,
30410, 10309, 100, 12670, 51609, 66232, 72622, 76022, 89428, 92727, 131314,
160023, 197924, 227711, 280067, 285519, 413874, 688813), direction = c(-1,
-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1), id = c("12", "12", "7.3", "7.3", "6.2",
"7.4", "2.3", "7.5", "2.3", "1", "7.4", "7.3", "7.1", "18", "7.1", "17",
"7.5", "X", "2.1", "7.4", "10", "6.2", "2.1", "1", "2.1", "11.1", "7.5",
"15.1", "15.1", "2.3", "15.2"), color = c("aquamarine3", "aquamarine3",
"aquamarine3", "aquamarine3", "orange1", "orange1", "lemonchiffon",
"goldenrod1", "lemonchiffon", "wheat4", "orange1", "aquamarine3",
"darkslategrey", "wheat4", "darkslategrey", "cyan", "goldenrod1",
"goldenrod1", "orange1", "orange1", "navajowhite1", "orange1", "orange1",
"wheat4", "orange1", "tomato", "goldenrod1", "navajowhite1",
"navajowhite1", "lemonchiffon", "darkred")), .Names = c("response",
"distance", "direction", "id", "color"), row.names = c(NA, -31L ), class =
"data.frame")

And I would like to fit a mixed effects model where the response is
my.df$response, the fixed effects are: as.numeric(my.df$distance) and
as.factor(my.df$direction), and the random effect isas.factor(my.df$id).

The attached plot (plot(x = my.df$distance*as.numeric(my.df$direction), y =
my.df$response, col = my.df$color, lwd = 2, xlab = "Direction", ylab =
"Response"))

gives me the sense that a generalized linear mixed effects model with a
Gamma family is appropriate, but I may be wrong.

Trying to fit such a model with:
library(lme4)
#set my.df$direction and my.df$id as factors
my.df$direction = as.factor(my.df$direction)
my.df$id = as.factor(my.df$id)
my.fit = glmer(response ~ distance + direction + (1|id), data = my.df,
family = Gamma())

I get many lines with a message of this sort:
(bG) nan @ pos 23: y= 0.549507; mu=-0.0772004; wt=1; y/mu=-7.11793;
log(y/mu) =1.#QNAN

And at the end this message:
Error: (maxstephalfit) PIRLS step-halvings failed to reduce deviance in
pwrssUpdate In addition: Warning message:
In checkScaleX(X, ctrl = control) : Some predictor variables are on very
different scales: consider rescaling

Are my data that messy that a glmer with a gamma family cannot be fitted to
them or am I doing something wrong?
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Rplot.png
Type: image/png
Size: 3669 bytes
Desc: not available
URL: <https://stat.ethz.ch/pipermail/r-sig-mixed-models/attachments/20140705/b3e06eb7/attachment.png>