Pass a vector of variable names to arrange() in dplyr
Hadley hasn't made this obvious in the help file--only in his NSE vignette. The versions of the functions followed by underscores use standard evaluation, so you pass them vectors of strings and the like.
If I understand your problem correctly, you can just replace arrange() with arrange_() and it will work.
Specifically, pass the vector of strings as the .dots argument when you do it.
> df %>% arrange_(.dots=c("var1","var3"))
var1 var2 var3 var4
1 1 i 5 i
2 1 x 7 w
3 1 h 8 e
4 2 b 5 f
5 2 t 5 b
6 2 w 7 h
7 3 s 6 d
8 3 f 8 e
9 4 c 5 y
10 4 o 8 c
========== Update March 2018 ==============
Using the standard evaluation versions in dplyr as I have shown here is now considered deprecated. You can read Hadley's programming vignette for the new way. Basically you will use !! to unquote one variable or !!! to unquote a vector of variables inside of arrange().
When you pass those columns, if they are bare, quote them using quo() for one variable or quos() for a vector. Don't use quotation marks. See the answer by Akrun.
If your columns are already strings, then make them names using rlang::sym() for a single column or rlang::syms() for a vector. See the answer by Christos. You can also use as.name() for a single column. Unfortunately as of this writing, the information on how to use rlang::sym() has not yet made it into the vignette I link to above (eventually it will be in the section on "variadic quasiquotation" according to his draft).
In the new version (soon to be released 0.6.0 of dplyr) we can make use of the quosures
library(dplyr)
vector_of_vars <- quos(var1, var3)
df %>%
arrange(!!! vector_of_vars)
# var1 var2 var3 var4
#1 1 i 5 i
#2 1 x 7 w
#3 1 h 8 e
#4 2 b 5 f
#5 2 t 5 b
#6 2 w 7 h
#7 3 s 6 d
#8 3 f 8 e
#9 4 c 5 y
#10 4 o 8 c
When there are more than one variable, we use quos and for a single variable it is quo. The quos will return a list of quoted variables and inside arrange, we unquote the list using !!! for evaluation