---
title: "Revised analysis in response to second set of PeerJ reviewer comments"
author: "A. Hauger and G. Roloff; Sep 9"
output: html_document
---

```{r, Packages, include = FALSE, message = FALSE}
if (!require('lme4')) install.packages('lme4'); library('lme4')
if (!require('ggplot2')) install.packages('ggplot2'); library('ggplot2')
if (!require('regclass')) install.packages('regclass'); library('regclass')
```


```{r, setup, message = FALSE, warning = FALSE}
setwd("~/PigeDNA/PeerJRevised.2")

AllData2 <- read.csv("AllData2.csv", header = TRUE)
summary(AllData2)
AllData2$Stream <- as.factor(AllData2$Stream)
AllData2$Season<-as.factor(AllData2$Season)
AllData2$Temperature <-as.numeric(as.character(AllData2$Temperature))
AllData2$Distance<- as.numeric(AllData2$Distance)

#Need to recode Distance so that the data are ordinal where 0 = 0m,
# 1 = 100m, 2 = 200m, 3 = 300m and 4 = 400m.

AllData2$DistanceRecode <- ifelse(AllData2$Distance==3,1,
           ifelse(AllData2$Distance==4,2,
                  ifelse(AllData2$Distance==5,3,
                         ifelse(AllData2$Distance==6,4,0))))

```

```{r, Global.Model}
# Per recommendation of reviewer, removed Season as a factor.
Global.Mod = glm(Detection ~ DistanceRecode + Velocity + Temperature + pH + Discharge
                 + Turbidity + factor(Stream), data=AllData2, family=binomial(link='logit'))
summary(Global.Mod)

# Use variance inflation factor to identify predictors that may be more related
# to other predictors than to the response variable. VIF > 5 are suspect.

VIF(Global.Mod)

# Remove discharge.

Global.Mod2 = glm(Detection ~ DistanceRecode + Velocity + Temperature + pH
                 + Turbidity + factor(Stream), data=AllData2, family=binomial(link='logit'))
summary(Global.Mod2)

VIF(Global.Mod2)

# Remove pH.

Global.Mod3 = glm(Detection ~ DistanceRecode + Velocity + Temperature
                 + Turbidity + factor(Stream), data=AllData2, family=binomial(link='logit'))
summary(Global.Mod3)

VIF(Global.Mod3)

```

```{r}
# Is interaction between distance and velocity potentially important?
DistVelTest <- glm(Detection ~ DistanceRecode * Velocity, data = AllData2, family = binomial(link='logit'))
summary(DistVelTest)
# The intercation does not appear to be important.
```

```{r, ConfInt}
round(confint(Global.Mod3),3) 
```

```{r, Plot, warning=FALSE}
ggplot(AllData2, aes(Temperature , Detection)) +
  stat_smooth(method="glm", method.args = list(family = "binomial"), formula=y~x,
              alpha=0.3, size=2) +
  geom_point(position=position_jitter(height=0.03, width=0)) +
  xlab("Temperature (C)") + ylab("Detection Probability") +
  coord_cartesian(ylim=c(0,1)) +
  theme(panel.background = element_blank()) +
  theme(axis.line = element_line(colour = "black")) + ggtitle("") +
  theme(axis.text.x = element_text(family="serif", size=12), axis.title.x = element_text(family="serif", size=12),
        axis.text.y = element_text(family="serif", size=12), axis.title.y = element_text(family="serif", size=12)) +
  theme(axis.title.y = element_text(margin = ggplot2::margin(t = 0, r = 10, b = 0, l = 0))) +
  theme(axis.title.x = element_text(margin = ggplot2::margin(t = 10, r = 0, b = 0, l = 0)))
```

```{r}
# Predict detection probability for min and max recorded temps.

TempPredict = predict(Global.Mod3, data.frame = AllData2, type = "response")
TempPredict

```