Datasets:

n8layman
/

herd-explorer-projections

inst_id stringclasses 884 values	year int16 2.03k 2.03k ⌀	draw int16 1 200 ⌀	value float32 0 1,401,954,910,785,167,400B ⌀
001002	2,025	1	5,946,508.5
001002	2,025	2	1,367,842.375
001002	2,025	3	1,021,611.5
001002	2,025	4	5,079,484.5
001002	2,025	5	3,160,593.75
001002	2,025	6	2,828,158.75
001002	2,025	7	1,254,327.625
001002	2,025	8	1,744,684.375
001002	2,025	9	1,876,536.25
001002	2,025	10	2,627,368
001002	2,025	11	2,249,423
001002	2,025	12	3,005,120.5
001002	2,025	13	4,228,009.5
001002	2,025	14	3,690,971
001002	2,025	15	3,995,494
001002	2,025	16	1,740,366.5
001002	2,025	17	1,311,042.375
001002	2,025	18	1,918,291.125
001002	2,025	19	2,038,941
001002	2,025	20	2,569,984
001002	2,025	21	1,177,135.625
001002	2,025	22	1,161,605.25
001002	2,025	23	1,264,128.375
001002	2,025	24	3,493,908.25
001002	2,025	25	2,391,511.25
001002	2,025	26	2,507,189.75
001002	2,025	27	5,682,504
001002	2,025	28	2,959,238.25
001002	2,025	29	1,145,882.125
001002	2,025	30	1,694,741.375
001002	2,025	31	2,228,367.75
001002	2,025	32	3,810,320.75
001002	2,025	33	1,498,899.75
001002	2,025	34	2,930,055.25
001002	2,025	35	3,100,597.5
001002	2,025	36	3,130,686
001002	2,025	37	1,866,952.375
001002	2,025	38	2,550,377.75
001002	2,025	39	3,190,977.5
001002	2,025	40	1,508,579.875
001002	2,025	41	5,102,537
001002	2,025	42	1,542,889
001002	2,025	43	2,698,691.75
001002	2,025	44	2,448,542
001002	2,025	45	2,202,458
001002	2,025	46	1,055,641.25
001002	2,025	47	1,173,845
001002	2,025	48	1,666,741.125
001002	2,025	49	1,440,838.75
001002	2,025	50	4,974,922
001002	2,025	51	2,768,933
001002	2,025	52	1,031,343.6875
001002	2,025	53	3,625,220
001002	2,025	54	2,376,038
001002	2,025	55	1,770,699.75
001002	2,025	56	951,430.625
001002	2,025	57	1,612,172.5
001002	2,025	58	1,546,386.125
001002	2,025	59	3,301,649.25
001002	2,025	60	2,368,902
001002	2,025	61	3,238,528.5
001002	2,025	62	2,858,535.5
001002	2,025	63	1,938,060.125
001002	2,025	64	5,863,981
001002	2,025	65	2,187,478
001002	2,025	66	3,485,042.25
001002	2,025	67	2,315,168.25
001002	2,025	68	1,648,377
001002	2,025	69	1,898,783.25
001002	2,025	70	1,185,387.375
001002	2,025	71	2,710,652.25
001002	2,025	72	2,265,860.75
001002	2,025	73	2,491,666
001002	2,025	74	1,564,723.25
001002	2,025	75	2,466,855
001002	2,025	76	2,182,643
001002	2,025	77	2,550,624.75
001002	2,025	78	3,453,360.5
001002	2,025	79	3,039,105.5
001002	2,025	80	1,922,848
001002	2,025	81	1,352,847.75
001002	2,025	82	2,172,620.5
001002	2,025	83	2,575,132.75
001002	2,025	84	1,675,099.125
001002	2,025	85	2,041,570.875
001002	2,025	86	4,413,866.5
001002	2,025	87	1,498,614.25
001002	2,025	88	1,366,171.25
001002	2,025	89	2,262,443.25
001002	2,025	90	2,250,245.25
001002	2,025	91	1,466,446
001002	2,025	92	2,362,970
001002	2,025	93	4,492,349.5
001002	2,025	94	1,206,335.625
001002	2,025	95	1,702,548
001002	2,025	96	4,663,020
001002	2,025	97	1,560,813.625
001002	2,025	98	1,972,753.625
001002	2,025	99	4,178,224
001002	2,025	100	3,938,662.75

End of preview. Expand in Data Studio

HERD Explorer projections

Simulated projections of research and development spending at U.S. colleges and universities, FY 2025 to FY 2029, for every dollar measure in HERD Explorer. The site reads these files directly in the browser.

A projection shows where past trends lead if they continue. It does not account for known upcoming changes, such as federal budget or policy shifts, and may not reflect what actually happens. Nor can it see survey revisions or an institution's own plans.

Files

<measure_id>.parquet: 200 simulated futures per institution and year. Columns: inst_id (NCSES HERD institution ID), year, draw (1 to 200), value (whole nominal dollars; NCSES publishes HERD in thousands, multiplied by 1,000 here). Rows are sorted by institution. To get a range for any group of institutions, add the institutions' values within each draw, then take percentiles across draws.
by_group/<measure_id>.parquet: the same draws already summed for every group the site shows (split is empty for the national total, or sector, control, state, msi, hbcu, medical_school, land_grant; grp is the group). Institutions the model did not project are included at their last reported value.
by_institution/inst_id=<id>/data_0.parquet: every measure's draws for one institution (measure_id, year, draw, value).
model_info.parquet: one row per measure, with whether it was modeled, the reason if not, and fitted parameters.
backtest*.parquet: how the model scored against past years (CRPS and interval coverage), overall, by institution size, and for each candidate interval width.

Measure definitions are in the measure_catalog table of the HERD Explorer database.

Method

A hierarchical growth model fit with the R package mgcv: year-to-year growth in log spending, with typical growth by institution type (2021 Carnegie Classification), state and institution random effects, shared year and year-by-type shocks, and noise that depends on institution size. Chosen over a naive baseline and a model of spending levels by rolling backtests. Full description: About the model. Code: model/fit.R.

Source data

NCSES Higher Education Research and Development Survey public microdata, FY 2012 to FY 2024.

Disclaimer

HERD Explorer is an independent project. It is not an official product of NCSES or the U.S. National Science Foundation, and neither NCSES, NSF nor the University of Idaho endorses it. These projections are provided as is, without warranty of any kind, and are not financial, budget, legal or investment advice. Report problems at https://github.com/nate-layman/herd-explorer/issues

Citation

Layman, N. HERD Explorer. University of Idaho, AI4RA. https://github.com/nate-layman/herd-explorer

This material is based upon work supported by the U.S. National Science Foundation under the GRANTED program, award 2427549.

Downloads last month: 101