rnsf

Unofficial package to interface with NSF API

It also has abstracts, dates, and more information for all grants up to April 11, 2026.

Installation

This package has a huge (>500 MB) R data file containing all the cached grant info (abstracts, money, institutions, etc.). I have stored it as a git large file storage file inside data. Thus, the usual approach to installing from github won’t work as the data won’t be loaded in the correct way. Instead, from a terminal or command line, on a computer with git and git-lfs installed, you will have to:

git clone https://github.com/bomeara/rnsf.git (note this is in your computer terminal, not an R session!)

and then use R CMD INSTALL rnsf to install the package (or, from within R, devtools::install('rnsf') or whatever the path is to the directory you have cloned).

If someone wants to make an easier way to do this, please reach out!

Examples

Getting cached data

This package caches data on over 500,000 funded NSF grants (it’s large, which is why the package will take a while to install, and why it will never be on CRAN with its 5 MB maximum size). You can use it.

library(rnsf)
data(grants)

This will give you a data.frame with all data: head(grants)

It has multiple date fields which are in “%m/%d/%Y” format (but everything is stored as a raw list, leaving to the user to process it).

You can also do a new search. For example, to get all info on ants:

library(rnsf)
ants <- rnsf::nsf_return(keyword="Formicidae")
#> [1] "Finished first batch"

Visualization

Wordcloud

There is a function for doing a wordcloud:

library(rnsf)
nsf_wordcloud(ants$abstractText)

Topic frequency over time

The ozone hole was discovered in 1985: pollution led to a depletion of the ozone layer, which shields people (and other organisms) from a great deal of UV radiation. The world came together and passed the Montreal Protocol to limit the pollution, chlorofluorocarbons, that was causing the damage. The hole is healing, but still requires research and monitoring. We can see how NSF grants mentioning “ozone hole” changed over time (though not all grants mentioning “ozone hole” study this issue; some could be using it as a comparison to some other issue, for example). We can include a regression before and after 1985 and show the 95% CI for the proportion in each year (truncated by the y-axis limits).

library(rnsf)
library(ggplot2)
library(timeDate)
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(viridis)
#> Loading required package: viridisLite
library(binom)

data(grants) # or use grants <- rnsf::nsf_get_all() to get the most current list (will take hours) or grants <- nsf_update_cached() to update from the last cached version

grants$ozone_hole <- grepl("ozone hole", grants$abstractText, ignore.case=TRUE)
grants$date <- as.Date(grants$date, format="%m/%d/%Y")
grants$year <- as.numeric(format(grants$date, "%Y"))

grants_by_year <- grants |> group_by(year) |> summarize(ozone = sum(ozone_hole), all = n())  |> ungroup()
grants_by_year$ozone_percent <- 100 * grants_by_year$ozone / grants_by_year$all
grants_by_year$ozone_lower_percent <- 100 * binom::binom.confint(x=grants_by_year$ozone, n=grants_by_year$all, method="exact")$lower
grants_by_year$ozone_upper_percent <- 100 * binom::binom.confint(x=grants_by_year$ozone, n=grants_by_year$all, method="exact")$upper
grants_by_year$ozone_upper_percent_truncated <- sapply(grants_by_year$ozone_upper_percent, min, 2*max(grants_by_year$ozone_percent))
g <- ggplot(grants_by_year, aes(x=year, y=ozone_percent, group=1))+ geom_errorbar(aes(ymin=ozone_lower_percent, ymax=ozone_upper_percent_truncated), colour="gray", alpha=0.4, width=0) + geom_point() + theme_minimal() + xlab("Year") + ylab("Percentage of grants mentioning 'ozone hole'") + geom_smooth(data=subset(grants_by_year, year<1985), se=FALSE, method="lm") + geom_smooth(data=subset(grants_by_year, year>=1985), se=FALSE, method="lm")  + ylim(c(0, 2*max(grants_by_year$ozone_percent)))
print(g)
#> `geom_smooth()` using formula = 'y ~ x'
#> `geom_smooth()` using formula = 'y ~ x'

Bergogram

We can plot funding over time, inspired by (but not endorsed by) Jeremy Berg’s graphs on federal funding (see https://jeremymberg.github.io/jeremyberg.github.io/). Also see https://grant-witness.us/funding_curves.html#nsf-graphs from Grant Witness.

grants$date <- as.Date(grants$date, format="%m/%d/%Y")
grants$year <- format(grants$date, "%Y")
grants$dayofyear <- timeDate::dayOfYear(timeDate::timeDate(grants$date))
grants$estimatedTotalAmt <- as.numeric(grants$estimatedTotalAmt)
grants <- grants |> arrange(year, dayofyear) |> group_by(year) |> mutate(estimatedTotalAmt_by_year = cumsum(estimatedTotalAmt)) |> ungroup()
grants <- grants[!is.na(grants$year),]
grants_recent <- subset(grants, year>=2020)
g <- ggplot(grants_recent, aes(x=dayofyear, y=estimatedTotalAmt_by_year/1e9, colour=year)) + geom_line() + theme_minimal() + xlab("Day of year") + ylab("Cumulative billions of dollars awarded") + scale_colour_viridis_d(option="turbo", begin=0.2)
print(g)

Table of award info

We can also look at a table with the number, not total money, of grants by state or territory by academic semester, for example (only including this year up to the last cache of the data).

library(tidyr)
library(knitr)

data(grants)
grants$awardeeStateCode <- toupper(grants$awardeeStateCode) # handle some early data that uses lowercase
grants$academic_semester <- rnsf::date_to_academic_semester(as.Date(grants$date, format="%m/%d/%Y"))
grants_aggregated <- grants |> filter(academic_semester %in% apply(expand.grid(c(2024:2026), c(" Fall", "  Spring")), 1, paste0, collapse="")) |> group_by(academic_semester, awardeeStateCode) |> summarise(total_awarded = n(), .groups = "drop_last") |> ungroup() |> tidyr::pivot_wider(names_from = academic_semester, values_from=total_awarded, values_fill=0) |> dplyr::arrange(desc(`2024 Fall`))
colnames(grants_aggregated) <- gsub("  ", " ", colnames(grants_aggregated))
colnames(grants_aggregated)[1] <- "Area"
grants_aggregated[,1] <- rnsf::abbreviation_to_state(unname(unlist(grants_aggregated[,1])))
knitr::kable(grants_aggregated)
Area 2024 Spring 2024 Fall 2025 Spring 2025 Fall 2026 Spring
California 508 752 361 586 102
New York 371 475 256 371 63
Texas 324 415 224 355 64
Massachusetts 327 410 189 305 31
Pennsylvania 243 282 165 233 48
Illinois 218 277 122 224 26
Virginia 161 227 89 139 24
Florida 184 225 105 179 42
Michigan 208 215 97 188 26
North Carolina 153 210 103 181 30
Colorado 111 183 73 146 14
Arizona 101 177 65 99 15
Georgia 137 171 78 147 16
Indiana 130 166 83 157 28
Maryland 118 147 79 129 36
New Jersey 123 146 90 138 20
Washington 97 144 68 106 19
Ohio 110 134 73 106 17
Tennessee 78 108 41 83 11
Alabama 63 105 51 74 20
Wisconsin 76 105 64 91 18
South Carolina 60 102 31 66 17
District of Columbia 51 101 42 55 13
Minnesota 79 101 47 62 9
Rhode Island 66 84 46 71 21
Oregon 68 83 41 65 8
Iowa 54 77 36 66 7
Louisiana 54 77 30 68 12
Utah 44 77 35 65 11
Missouri 74 76 64 63 18
Connecticut 66 63 52 66 8
New Mexico 34 62 21 38 10
Oklahoma 42 59 23 51 10
Kansas 29 58 14 34 11
Nebraska 39 55 22 44 3
Hawaii 15 53 16 28 7
Kentucky 36 51 28 40 5
Delaware 18 47 31 37 6
Idaho 16 43 16 31 5
Nevada 12 43 11 31 3
New Hampshire 21 36 21 35 2
Mississippi 38 35 19 35 3
Alaska 10 32 4 19 2
Montana 19 32 8 22 3
Maine 30 27 10 24 2
Arkansas 27 25 10 26 6
West Virginia 28 25 8 26 1
South Dakota 20 22 17 17 1
Vermont 13 22 12 13 4
Puerto Rico 7 20 9 14 0
Wyoming 10 19 7 18 3
North Dakota 11 12 9 21 3
Virgin Islands of the U.S. 0 3 1 1 0
American Samoa 0 1 0 0 0
Guam 0 0 2 1 0

Recent info

We can see when the most recent grant has been awarded (relative to when this package was last built) by state or territory; we can also look to see how funding so far this year compares to average funding at this point of the year for 2017-2024 (so it encompasses two different administrations).

library(lubridate)
#> 
#> Attaching package: 'lubridate'
#> The following objects are masked from 'package:base':
#> 
#>     date, intersect, setdiff, union
library(ggrepel)

data(grants)

current_day_of_year <- lubridate::yday(Sys.Date())
grants_modified <- grants
grants_modified$awardeeStateCode <- toupper(grants_modified$awardeeStateCode) # handle some early data that uses lowercase
grants_modified$date_formatted <- as.Date(grants_modified$date, format="%m/%d/%Y")
grants_modified <- grants_modified[!is.na(grants_modified$date_formatted),]
grants_modified$day_of_year <- lubridate::yday(grants_modified$date_formatted)
grants_modified$year <- as.numeric(format(grants_modified$date_formatted, "%Y"))
grants_modified$funds <- as.numeric(grants_modified$fundsObligatedAmt)
grants_recent <- grants_modified |> group_by(awardeeStateCode) |> summarize(most_recent = max(date_formatted), days_since_grant = Sys.Date() - max(date_formatted))

grants_2026 <- subset(grants_modified, year==2026)
grants_2017_2024 <- subset(grants_modified, year>= 2017 & year <=2024) # so we have 8 years of data across two admins
grants_2017_2024 <- subset(grants_2017_2024, day_of_year < current_day_of_year) # so we compare similar
grants_aggregated_2026 <- grants_2026 |> group_by(awardeeStateCode) |> summarize(funds_per_year_2026 = sum(funds)) |> arrange(awardeeStateCode) |> ungroup()
grants_aggregated_2017_2024 <- grants_2017_2024 |> group_by(awardeeStateCode) |> summarize(funds_per_year_8 = sum(funds)/8) |> arrange(awardeeStateCode) |> ungroup()
grants_comparison <- full_join(full_join(grants_aggregated_2017_2024, grants_aggregated_2026), grants_recent)
#> Joining with `by = join_by(awardeeStateCode)`
#> Joining with `by = join_by(awardeeStateCode)`
grants_comparison$funds_per_year_2026[is.na(grants_comparison$funds_per_year_2026)] <- 0
grants_comparison <- subset(grants_comparison, !is.na(grants_comparison$funds_per_year_8)) # rarely get funding
grants_comparison <- subset(grants_comparison, grants_comparison$funds_per_year_8>50000) # eliminate entities which have traditionally low, rare funding as they make it hard to see the wait times for the others 
grants_comparison$percentage_usual_funding_by_this_date <- 100*grants_comparison$funds_per_year_2026 / grants_comparison$funds_per_year_8
grants_comparison$days_since_grant <- as.numeric(grants_comparison$days_since_grant)
g <- ggplot(grants_comparison, aes(x=days_since_grant, y=percentage_usual_funding_by_this_date, label=awardeeStateCode)) + geom_text_repel(max.overlaps=50, alpha=0.8) + theme_minimal() + xlab("Days since most recent grant") + ylab("2026 percentage of usual funding by this date") + geom_point(alpha=0.5) + scale_x_continuous(trans='log1p', breaks=c(1, 7, 30, 90, 120, 365, 2*365), limits=c(0, NA)) 
print(g)

GRFP data

The NSF Graduate Research Fellowship Program (GRFP) is one of NSF’s oldest and most impactful programs: it provides funding for three years of study for people in grad school, giving them flexibility (no need for a research or teaching assistantship); the stipends are also often higher than those for most grad students. Every year, names, affiliations, and research areas of awardees (those receiving the money) and those with honorable mentions are released here but you can only get one year at a time. I have manually downloaded them all and incorporated them into the package. To use:

library(rnsf)
data(grfp)

You can then plot information or do other analyses. For example, the number of awards in Mathematical Sciences over time:

grfp_math <- grfp[grepl("Mathematical Sciences", grfp$Field.of.Study),] |> subset(award_level=="Awardee")
grfp_math$year <- as.numeric(grfp_math$year)
grfp_math_by_year <- grfp_math |> group_by(year) |> summarize(count=n()) |> ungroup()
g <- ggplot(grfp_math_by_year, aes(x=year, y=count, group=1)) + geom_line() + theme_minimal() + xlab("GRFP Year") + ylab("Number of awards in 'Mathematical Sciences'")
print(g)

And the frequency of different subfields of math, showing just the first twenty from the past ten years of awards:

library(stringr)
subfields <- stringr::str_to_title(gsub("Mathematical Sciences - ", "", grfp_math$Field.of.Study))
popularity <- t(t(sort(table(subfields), decreasing=TRUE)))
knitr::kable(head(popularity, 20))
Algebra Or Number Theory 786
Analysis 758
Mathematical Sciences 539
Topology 466
Applications Of Mathematics (Including Biometrics And Bi 465
Algebra, Number Theory, And Combinatorics 359
Applied Mathematics 239
Probability And Statistics 227
Logic Or Foundations Of Mathematics 172
Geometry 139
Statistics 131
Mathematical Biology 84
Operations Research 76
Biostatistics 70
Computational Mathematics 51
Geometric Analysis 37
Probability 31
Computational And Data-Enabled Science 25
Artificial Intelligence 16
Computational Statistics 16

Updating the package

First, update the package version in the DESCRIPTION.

Then the directory containing the package source:

library(rnsf)
grants <- nsf_update_cached_this_year() # perhaps worth doing a new nsf_get_all() after the beginning of the year
usethis::use_data(grants, overwrite=TRUE) 
devtools::install()

Then quit R and reopen (so it uses the latest saved grants).

Then:

library(rnsf)
devtools::build_readme()
pkgdown::build_site()
system("git add */*")
system("git commit -m'automatic update of page and data' -a")
system("git push")

If new GRFP results are out, download them, retitle them as Awardee or HonorableMention with year and .tsv (i.e., “Awardee2030.tsv”) and put them in the inst/extdata directory in the package source. Reinstall the package. Then,

library(rnsf)
grfp <- compile_grfp()
usethis::use_data(grfp, overwrite=TRUE)

Then do the usual git things.