Calculate moving geometric mean by water sampling station - r

Question

I need to calculate the moving geometric mean on fecal coliform over time(at each value I want the geomean of that value and the previous 29 values), by individual sampling stations. When I download the data from our database the column headers are:
Station SampleDate FecalColiform
Depending on the growing area there are a few to over a dozen stations.
I tried to adapt some code that I found at HERE:
#File: Fecal
Fecal <- group_by(Fecal, Station) %>%
arrange(SampleDate) %>%
mutate(logres = log10(ResultValue)) %>%
mutate(mgm = stats::filter(logres, rep(1/24, 24), sides =1))
This worked, but the problem is that I don't want the resulting log values. I want just the regular geomean so that I can plot it and everyone can easily understand the values. I tried to somehow sneak the geometric.mean function from the psych package in there I could not make that work.
There are resources for calculating a moving average, and code for calculating geometric mean and I have tried to combine several of them. I can't find an example for moving geometric mean.
Eventually I would like to graph all of geomeans by station similar to the example in the link above.
> dput(ByStationRGMData[1:10,])
structure(list(Station = c(114L, 114L, 114L, 114L, 114L, 114L,
114L, 114L, 114L, 114L), Classification = structure(c(3L, 3L,
3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L), .Label = c(" Approved ", " Conditionally Approved ",
" Prohibited "), class = "factor"), SampleDate = c(19890103L,
19890103L, 19890209L, 19890316L, 19890413L, 19890511L, 19890615L,
19890713L, 19890817L, 19890914L), SWTemp = c(NA, NA, 5L, 8L,
NA, 13L, 15L, 18L, NA, 18L), Salinity = c(NA, NA, 22L, 18L, NA,
26L, 22L, 24L, NA, 32L), FecalColiform = c(180, 49, 2, 17, 7.9,
1.8, 4.5, 11, 33, 1.8), RGM = c(NA_real_, NA_real_, NA_real_,
NA_real_, NA_real_, NA_real_, NA_real_, NA_real_, NA_real_, NA_real_
)), .Names = c("Station", "Classification", "SampleDate", "SWTemp",
"Salinity", "FecalColiform", "RGM"), class = c("grouped_df",
"tbl_df", "tbl", "data.frame"), row.names = c(NA, -10L), vars = list(
Station), drop = TRUE, indices = list(0:9), group_sizes = 10L, biggest_group_size = 10L, labels = structure(list(
Station = 114L), class = "data.frame", row.names = c(NA,
-1L), vars = list(Station), drop = TRUE, .Names = "Station"))
I would also like to add a moving 90th percentile to the dataframe and the graphs. I tried the following:
ByStationRGMData <- RawData %>%
group_by(Station) %>%
arrange(SampleDate) %>%
mutate(RGM = as.numeric(rollapply(FecalColiform, 30, geometric.mean, fill=NA, align="right"))) +
mutate(F90 = as.numeric(rollapply(FecalColiform, 30, quantile, p=0.90, fill=NA, align="right")))
This gives me the error:
Error in mutate_(.data, .dots = lazyeval::lazy_dots(...)) : argument ".data" is missing, with no default
I can't seem to figure out what I'm missing.

Answer 1

You can use rollapply from the zoo package (illustrated here using the built-in mtcars data frame). I've used a window of 3 values, but you can set that to 30 in your actual data. align="left" uses the current value and n-1 previous values, where n is the window width:
library(psych)
library(dplyr)
library(zoo)
mtcars %>%
mutate(mpgGM = rollapply(mpg, 3, geometric.mean, fill=NA, align="left"))
Include a grouping variable to get rolling geometric means separately for each group.