I am running a script to find out differences between songs of birds (comparing different lengths, frequencies and others). I am using linear mixed effects with lme4 package. I get as an outcome of negative Estimate Std. and since (for instance) the length of the song can not be negative, I wonder if anybody could tell me what I am doing wrong. Find details underneath.
I have been looking for errors in my data and different ways to dispose of the data, getting the same results.
This is how I have the data organized:
Bird site length freq
1 FH 2.69 4354 -58.9
1 FH 2.546 4298 -57.3
1 FH 2.043 5303 -53.7
2 FH 4.437 6084 -63.1
11 ML 3.371 4689 -37.1
12 ML 3.706 5470 -39.7
13 ML 4.331 5358 -48.7
13 ML 4.124 4744 -39.8
14 ML 3.802 5805 -42.5
This is the full code
#1 song lenght####
library("lmerTest")
model1<-lmer(length~site
+(1|Bird),
data=dframe1)
summary(model1)
anova(model1, test="F")
pdat <- expand.grid (site=c("ML", "SI","FH", "SH"))
detach(package:lmerTest) #
model1<-lmer(length~site
+(1|Bird),
data=dframe1)
pred <- predictSE(model1, newdata = pdat, re.form = NA,
se.fit = T, na.action = na.exclude,
type= "response")
pred
predframe <- data.frame (pdat, pred) ; predframe
predframe
plot(
NULL
, xlim = c(0.75,4.25) #
, ylim = c(3,6)
, axes = F #
, ylab = ""
, xlab = ""
)
at.x <- c(1,2,3,4)
at.lab <- c(1,2,3,4)
for (i in 1:nrow(predframe))
{arrows(
x0 = at.x[i]
, y0 = (predframe$fit[i] + predframe$se.fit[i])
, x1 = at.x[i]
, y1 = (predframe$fit[i] - predframe$se.fit[i])
, code = 3
, angle = 90
, length = 0.12
, col = "gray25")
points(
x = at.x[i]
, y = predframe$fit[i]
, pch = 21
,bg="black"
, col = "black"
, cex = 1.25) # point size
}
axis(1, labels = c("Mainland","Sully", "Flat Holm","Skokholm"), at = at.lab)
axis(2, at = c(3,4,5,6), labels = c(3,4,5,6), las = 1, cex.axis = 1)
box()
title(xlab = "Location", line = 2.5, cex = 0.8)
title(ylab = expression(paste("song length (secs)")), line = 2.75)
Ahead is the first part of the results, not sure why the site FH (siteFH -0.9480) comes up as negative. This happens with other variables as well, so I guess must be something wrong with the model. I am a beginner, please be considered with me, I've looked already and I haven't found a similar question.
Thank you in advance.
Results
`Scaled residuals:
Min 1Q Median 3Q Max
-3.1852 -0.4119 -0.0071 0.5304 2.2659
Random effects:
Groups Name Variance Std.Dev.
Bird (Intercept) 0.51798 0.7197
Residual 0.07313 0.2704
Number of obs: 112, groups: Bird, 42
Fixed effects:
Estimate Std. Error df t value Pr(>|t|)
(Intercept) 4.2429 0.1787 37.6710 23.745 < 2e-16 ***
siteFH -0.9480 0.2965 36.3879 -3.197 0.002871 **
siteSH 1.2641 0.3173 35.4150 3.983 0.000323 ***
siteSI -0.4258 0.3515 35.2203 -1.212 0.233769
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Correlation of Fixed Effects:
(Intr) siteFH siteSH
siteFH -0.603
siteSH -0.563 0.339
siteSI -0.508 0.306 0.286
> anova(model1, test="F")
Type III Analysis of Variance Table with Satterthwaite's method
Sum Sq Mean Sq NumDF DenDF F value Pr(>F)
site 3.0075 1.0025 3 35.336 13.709 4.337e-06 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1`
The columns in the output are right aligned, so the column is named Estimate, the next column is named Std. Error.
The estimate describes the association between your dependent and independent variables. It does not describe any values in your dataset.
A negative estimate just means "the larger your dependent variable (length), the lower your independent variable (site)" (or vice versa). But within this relationship, both variables still can be positive.
In detail, an estimate of -0.948 in your case means that the length for siteFH is about 0.948 lower than the length for siteML (the reference category, not shown in the output). However, it does not mean that siteFH is negative.