Does the "R" method for creating computed variables use a kind of pseudo R coding?
When I use the "R" option for creating a computed variable, I see that I don't use normal R methods to refer to variables within data frame. For example, with the Tooth Growth data set, if I create a variable called "z" and then make z equal to the square root of the variable "len," I specify "sqrt(len") instead of the usual R methods such as "sqrt(data[,"len"])" or "sqrt(data$len)". So does jasp's R option (for computed variables) use what might be called a "pseudo R" coding? Or, perhaps, is there some special R package being employed?
R
Comments
Actually, I meant to indicate something like "sqrt(data[i,"len"]" within an i loop.
R
I'll ask our expert. Sorry for the tardy response, I stopped getting emails from the Forum for some reason.
It's not just "pseudo R" coding, it's just plain R code with some restrictions (for example, some functions are not allowed). We do make some custom functions available, like z-score transformations, but in principle, with the R option you can write your own R code to create a column or filter. The last statement is used as the "returned value", so it's also fine to split code over multiple lines.
Still, regarding the following data frame:
mydata = data.frame(
person = c(1, 2, 3),
q = c(10, 100, 1000),
r = c(20, NA, 2000),
s = c(30, 300, 3000)
)
In base R, one way to get a mean for each person is . . .
mydata$MyNewVar2 <- NA
for (i in c(1:nrow(data))) {
mydata[i, "MyNewVar2"] <- mean(as.numeric(mydata[i,c("q", "r", "s")]), na.rm = TRUE)
}
Another way is as follows . . .
mydata$MyNewVar3 <- apply(mydata[, c("q", "r", "s")], 1, mean, na.rm = TRUE)
Using dplyr, one can write . . .
mydata = mutate(rowwise(mydata), MyNewVar = mean(c(q, r, s), na.rm = TRUE))
It seems that jamovi "R" syntax is closest to the of dplyr R. Is it exactly dplyr R, or is it something else? (It definitely is not base R.)
Thanks.
R
I'm not sure what jamovi does, but in R we simply execute the R code and try to use the return value.
So if this is the dataset loaded in JASP:
then doing
MyNewVar2 <- rep(0, length(q)) for (i in c(1:nrow(data))) { MyNewVar2 <- mean(as.numeric(c(q[i], r[i], s[i])]), na.rm = TRUE) } MyNewVar2is equivalent to what you wrote above. Here is a screenshot:
You can do the same with apply, albeit that we don't make a dataset object available so you'd have to create your own first:
Hope that helps!
Thanks.
One thing I notice is that with your R code, you are creating a new vector rather than a new data-frame column (it was not obvious to me that JASP only wanted a new vector).
However, it's still the case that if I use drag and drop to compute (q + r + s) / 3, JASP informs me that the corresponding R code is "(q + r + s) / 3" -- no looping or anything like that! (See the screen shot below.)
Moreover, if I then make use of JASP's option to use "R" (rather than drag-and-drop) to recreate the same variable (this time as MyNewVar3 instead of MyNewVar2) -- and if I simply enter the "R" code, (q + r + s) / 3 -- it works! (See the screen shot, below.) Yet there is no looping, no na.rm, no nothing (except (q + r + s) / 3).
So what's the explanation for this?
R
Well, the drag and drop is under the hood translated to R code. If you want, this can be shown to you. In R, many operations (e.g., +, -, *, /) are vectorized if the vectors have the same length. For example, you can add two vectors. So to compute the rowwise mean you can indeed do
albeit that missing values (NA) are propagated.
OK. Thanks. Here are the conclusions I've drawn with regard to instructing students and others on the use of R to compute new variables in JASP:
"
The JASP data set is accessed, *not* as an R data frame, but as a collection of *individual vectors*. Thus for example, if there's an existing JASP data column, x, and you want to create a new column in which each new value equals x + 5, you can use the expression: x + 5
Because the x variable you interact with is *not* part of a data frame, you should not attempt anything resembling: df[ , "x"] + 5
That said, if you use happen to have used R computations within JASP to create your own R data frame (which, to re-emphasize, won't constitute a JASP data set), then you can manipulate the data frame's elements the way you normally would in R.
"
R