Tailflip as a Predictor
Jesi
2025-07-05
This document is my attempt to determine whether my quantification of nociception predicted avoidance learning via time in each zone (dark, light, middle)
data<-Behavioral.Trials.All.DataVariable Transformations
#LIGHT
#Make new column with no 0 values
min_val <- min(data$light)
data$light_shift <- data$light + abs(min_val) + 1
#Transform
lambda <- 0.1262881
data$light_trans <- data$light_shift ^ lambda
#MIDDLE
#make new column with no 0 values
min_val <- min(data$middle)
data$mid_shift <- data$middle + abs(min_val) + 1
#Transform
lambda <- 0.06642153
data$mid_trans <- data$mid_shift ^ lambda
#DARK
#dark is normal enough. Transformation below increases skewness.
#make new column with no 0 values
min_val <- min(data$dark)
data$dark_shift <- data$dark + abs(min_val) + 1
#Transform
lambda <- 0.8567983
data$dark_trans <- data$dark_shift ^ lambda#Separate trials
T1<- data[data$trial=="1",]
T2<- data[data$trial=="2",]
T3<- data[data$trial=="3",]
data23<- rbind(T2,T3)
#separate experiments
exp1<-(data[data$exp=="1",])
exp2<-(data[data$exp=="2",])
#separate treatment groups
hot<-(data[data$hot=="y",])
control<-(data[hot$exp=="n",])Plots
ggplot(data, aes(x = tailflip, y = light_trans)) +
geom_jitter(width = 0.1, height = 0, alpha = 0.6) +
geom_smooth(method = "lm", se = TRUE, color = "black") +
theme_minimal() +
labs(title = "Tailflips vs. Transformed Time in Light",
x = "Tailflips per Trial", y = "Transformed Time in Light")## `geom_smooth()` using formula = 'y ~ x'
ggplot(data, aes(x = tailflip, y = mid_trans)) +
geom_jitter(width = 0.1, height = 0, alpha = 0.6) +
geom_smooth(method = "lm", se = TRUE, color = "black") +
theme_minimal() +
labs(title = "Tailflips vs. Transformed Time in Middle",
x = "Tailflips per Trial", y = "Transformed Time in Middle")## `geom_smooth()` using formula = 'y ~ x'
ggplot(data, aes(x = tailflip, y = dark)) +
geom_jitter(width = 0.1, height = 0, alpha = 0.6) +
geom_smooth(method = "lm", se = TRUE, color = "black") +
theme_minimal() +
labs(title = "Tailflips vs. Transformed Time in Dark",
x = "Tailflips per Trial", y = "Time in Dark")## `geom_smooth()` using formula = 'y ~ x'
#AS YOU CAN SEE BELOW, LIGHT HAD WEAK POSITIVE TREND ASSOCIATED WITH TAILFLIPS AND DARK HAD WEAK NEGATIVE TREND ASSOCIATED WITH TAILFLIPS
Modelling
#IN ALL MODELS, INTERACTION TERM TAILFLIP * HOT WAS NOT SIGNIFICANT, DID NOT IMPROVE THE MODEL
Is tailflip predictive of Light?
# Fit linear mixed-effects model
model_tailflip_light <- lmer(light_trans ~ tailflip + (1 | subject) + (1 | trial), data = data)
#interaction with hot here is non significant and does not improve the model.
# View model summary
summary(model_tailflip_light)## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: light_trans ~ tailflip + (1 | subject) + (1 | trial)
## Data: data
##
## REML criterion at convergence: 121.9
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.2388 -0.7377 0.1322 0.5938 2.5073
##
## Random effects:
## Groups Name Variance Std.Dev.
## subject (Intercept) 0.0535258 0.23136
## trial (Intercept) 0.0007442 0.02728
## Residual 0.1046467 0.32349
## Number of obs: 138, groups: subject, 23; trial, 3
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 1.50908 0.06197 15.49571 24.353 9e-14 ***
## tailflip 0.01638 0.01208 123.96213 1.356 0.178
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## tailflip -0.363#weak positive trend
#Checking Residuals
plot(model_tailflip_light) # Residuals vs. fitted
qqnorm(resid(model_tailflip_light)) # Q-Q plot
qqline(resid(model_tailflip_light))
#residuals look good
#Testing whether ANY tail flipping predicts light:
data$tailflip_binary <- ifelse(data$tailflip > 0, 1, 0)
model_tailflip_binary <- lmer(light_trans ~ tailflip_binary + (1 | subject) + (1 | trial), data = data)
summary(model_tailflip_binary)## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: light_trans ~ tailflip_binary + (1 | subject) + (1 | trial)
## Data: data
##
## REML criterion at convergence: 117.7
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.3604 -0.7369 0.1426 0.6083 2.5034
##
## Random effects:
## Groups Name Variance Std.Dev.
## subject (Intercept) 0.050695 0.22516
## trial (Intercept) 0.001103 0.03321
## Residual 0.104377 0.32307
## Number of obs: 138, groups: subject, 23; trial, 3
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 1.48329 0.06660 18.63758 22.271 6.9e-15 ***
## tailflip_binary 0.10091 0.05964 124.34091 1.692 0.0932 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## talflp_bnry -0.500#weak positive estimate.
#Trials with tailflips spent 0.09 transformed units more time in the light. Not significant
Is tailflip predictive of Dark?
# Fit linear mixed-effects model
model_tailflip_dark <- lmer(dark ~ tailflip + (1 | subject), data = data)
#interaction here is not significant and greatly increases uncertainty
#trial had 0 variance in this model, took it out
# View model summary
summary(model_tailflip_dark)## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: dark ~ tailflip + (1 | subject)
## Data: data
##
## REML criterion at convergence: 1822.3
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.5758 -0.4027 0.1893 0.6492 1.8485
##
## Random effects:
## Groups Name Variance Std.Dev.
## subject (Intercept) 7215 84.94
## Residual 30778 175.44
## Number of obs: 138, groups: subject, 23
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 413.006 26.079 33.270 15.837 <2e-16 ***
## tailflip -7.323 6.429 131.057 -1.139 0.257
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## tailflip -0.459#no - weak negative trend
#Each additional tailflip weakly predicts ~7 units less time in the dark
#Checking Residuals
plot(model_tailflip_dark) # Residuals vs. fitted
qqnorm(resid(model_tailflip_dark)) # Q-Q plot
qqline(resid(model_tailflip_dark))
#residuals look pretty good
# Testing whether ANY tail flipping predicts dark:
model_tailflip_binaryd <- lmer(dark ~ tailflip_binary + (1 | subject) + (1 | trial), data = data)## boundary (singular) fit: see help('isSingular')summary(model_tailflip_binaryd)## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: dark ~ tailflip_binary + (1 | subject) + (1 | trial)
## Data: data
##
## REML criterion at convergence: 1820.2
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.5585 -0.4248 0.1975 0.6773 1.6283
##
## Random effects:
## Groups Name Variance Std.Dev.
## subject (Intercept) 7.097e+03 8.424e+01
## trial (Intercept) 3.502e-11 5.918e-06
## Residual 3.111e+04 1.764e+02
## Number of obs: 138, groups: subject, 23; trial, 3
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 391.87 29.19 48.13 13.43 <2e-16 ***
## tailflip_binary 13.43 31.96 130.57 0.42 0.675
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## talflp_bnry -0.611
## optimizer (nloptwrap) convergence code: 0 (OK)
## boundary (singular) fit: see help('isSingular')#not at all
Is tailflip predictive of Middle?
# Fit linear mixed-effects model
model_tailflip_mid <- lmer(mid_trans ~ tailflip + (1 | subject) + (1 | trial), data = data)
# View model summary
summary(model_tailflip_mid)## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: mid_trans ~ tailflip + (1 | subject) + (1 | trial)
## Data: data
##
## REML criterion at convergence: -102.5
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -1.89162 -0.60076 0.03973 0.62856 1.97621
##
## Random effects:
## Groups Name Variance Std.Dev.
## subject (Intercept) 0.002297 0.04793
## trial (Intercept) 0.001005 0.03171
## Residual 0.023082 0.15193
## Number of obs: 138, groups: subject, 23; trial, 3
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 1.249e+00 2.656e-02 3.830e+00 47.045 1.96e-06 ***
## tailflip 6.432e-04 5.451e-03 1.331e+02 0.118 0.906
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## tailflip -0.382#not at all
#Checking Residuals
plot(model_tailflip_mid) # Residuals vs. fitted
qqnorm(resid(model_tailflip_mid)) # Q-Q plot
qqline(resid(model_tailflip_mid))
#residuals look good
# Testing whether ANY tail flipping predicts middle:
model_tailflip_binarym <- lmer(mid_trans ~ tailflip_binary + (1 | subject)+ (1 | trial), data = data)
summary(model_tailflip_binarym)## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: mid_trans ~ tailflip_binary + (1 | subject) + (1 | trial)
## Data: data
##
## REML criterion at convergence: -106.8
##
## Scaled residuals:
## Min 1Q Median 3Q Max
## -2.00295 -0.62452 0.05746 0.62950 1.99137
##
## Random effects:
## Groups Name Variance Std.Dev.
## subject (Intercept) 0.002281 0.04776
## trial (Intercept) 0.001115 0.03339
## Residual 0.022876 0.15125
## Number of obs: 138, groups: subject, 23; trial, 3
##
## Fixed effects:
## Estimate Std. Error df t value Pr(>|t|)
## (Intercept) 1.26618 0.02936 4.98434 43.120 1.32e-07 ***
## tailflip_binary -0.02794 0.02693 133.01195 -1.038 0.301
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Correlation of Fixed Effects:
## (Intr)
## talflp_bnry -0.512#no. negative estimate