does pdf work with travis now?

.travis.yml

@@ -49,7 +49,7 @@ r_github_packages:
 script:
     - |
         Rscript -e 'bookdown::render_book("index.Rmd", "bookdown::gitbook")'
-#        Rscript -e 'bookdown::render_book("index.Rmd", "bookdown::pdf_book")'
+        Rscript -e 'bookdown::render_book("index.Rmd", "bookdown::pdf_book")'
 
 deploy:
     provider: pages

linear-models.Rmd

@@ -61,8 +61,8 @@ size) and `wt` (car weight):
 
 ```{r}
 mtcars %>%
-	select(mpg, disp, wt) %>%
-	pairs
+  select(mpg, disp, wt) %>%
+  pairs
 ```
 
 Before running any model, we should ask outselves: "what question we are trying
@@ -579,9 +579,9 @@ inputs.
 
 However, be cautious when standardising. You will sometimes see people interpret
 standardised coefficients in terms of 'relative importance' of the predictors.
-For example, they might say that if $\beta^1 = .2$ and $\beta^2 = .4$ then
-$\beta^2$ is twice as important as $\beta^$. Although this is appealing, it's
-not always valid.
+For example, they might say that if $\beta^{1}= .2$ and $\beta^{2} = .4$ then
+$\beta^{2}$ is twice as important as $\beta^{1}$. Although this is appealing,
+it's not always valid.
 
 The main problem is that you don't always know whether you have a full range of
 values of predictors in your sample. For example, imagine a case where the a
@@ -591,9 +591,9 @@ general population.
 ```{r, include=F}
 set.seed(1234)
 incomes <- data_frame(age = round(rnorm(1000, 40, 15)),
-	gender = sample(0:1, 1000, replace=T),
-	income = round(rnorm(1000,  30000, 1000) + 30*age + 1000 * gender)) %>%
-		mutate(gender=factor(gender, labels=c("F", "M")))
+  gender = sample(0:1, 1000, replace=T),
+  income = round(rnorm(1000,  30000, 1000) + 30*age + 1000 * gender)) %>%
+    mutate(gender=factor(gender, labels=c("F", "M")))
 ```
 
 We can plot these data to show the effect of age, and gender:
@@ -627,7 +627,7 @@ constant across the range of ages:
 
 ```{r}
 younger.incomes <- incomes %>%
-	filter(age<40)
+  filter(age<40)
 
 m2 <- lm(income~age+gender, data=younger.incomes)
 coef(m2)

link-functions.Rmd

@@ -79,10 +79,10 @@ You can think of the solution as coming in two steps:
 
 #### Step 1 {-}
 
-We can transform a probability on the 0---1 scale to a 0 \rightarrow ∞ scale by
-converting it to _odds_, which are expressed as a ratio:
+We can transform a probability on the 0---1 scale to a $0 \rightarrow ∞$ scale
+by converting it to _odds_, which are expressed as a ratio:
 
-$$\textrm{odds} = \dfrac{\textrm{probability}}{1-\textrm{probability}}$$
+$$odds = \dfrac{p}{1-p}$$
 
 Probabilities and odds ratios are two _equivalent_ ways of expressing the same
 idea.

power-analysis.Rmd

@@ -71,8 +71,11 @@ We now use the `powerSim` function to use our tweaked model to:
 -   count how many times (i.e., for what proportion) we get a significant _p_
     value
 
-```{r}
+```{r include=F, echo=T}
 power.sim <- powerSim(model.of.interest, nsim=100)
+```
+
+```{r}
 power.sim
 ```
 
@@ -82,7 +85,13 @@ to see what power we have to detect an effect of x = .8:
 
 ```{r}
 fixef(model.of.interest)['x'] <- .8
+```
+
+```{r include=F, echo=T}
 power.sim <- powerSim(model.of.interest, nsim=100)
+```
+
+```{r}
 power.sim
 ```
 
@@ -98,7 +107,13 @@ And we could simulate increasing the variance parameter for `g` to 0.5:
 
 ```{r}
 VarCorr(model.of.interest)['g'] <- .5
+```
+
+```{r include=F, echo=T}
 power.sim <- powerSim(model.of.interest, nsim=100)
+```
+
+```{r}
 power.sim
 ```