Fig 1.
This figure shows spending per month for R01 project periods that last three, four, five, or six years, relative to spending in the first month of the project period.
Estimates are from a regression of total expenditure (arcsinh-transformed) on a set of dummies for each month in a project period with project period fixed effects, with the first month of the project period as the excluded category. Separate regressions are run by project period length. Standard errors are clustered by expiring R01 project period.
Fig 2.
This figure compares the characteristics of expiring (and eventually renewed) R01 project periods that are used in the UMETRICS analysis.
Each unit of observation is an R01 project period. Fig 1A shows the number of R01s that were renewed within 30 calendar days of their expiry. Fig 1B is a histogram (30-day bins) for the number of days till renewal for R01s not renewed within 30 days. Fig 1C shows the smoothed density for total funding in the expiring project period for interrupted and uninterrupted R01s. Fig 1D shows the same figure for funding per year (total funding / length of expiring R01 project period). Fig 1E shows the proportion of projects by the expiring project period’s length.
Fig 3.
This figure shows event-study estimates (with 95% confidence intervals) of the difference in spending between PIs of interrupted and uninterrupted R01s.
Treated and control groups are matched on length of the expiring R01 project period and the regression is weighted using the matching weights. Each panel shows the estimates for a different outcome variable: total expenditure by PI (A), total vendor expenditure by PI (B), total labor expenditure by PI (C), and total number of employees paid by PI (D). Month 0 is the month that the focal R01 expires. Month -11 is the excluded category for the regression. Regressions are run separately on subsamples of PIs that have one R01 grant (green) or multiple R01s (brown), including R01-equivalents and P01 grants. Standard errors are clustered by expiring R01 project period.
Fig 4.
This figure shows event study coefficients with 95% confidence intervals (clustered by expiring R01 project period) of the difference in probability of being paid for employees on an interrupted project relative to those on an uninterrupted project.
The same event study is estimated on subsamples by occupation and number of R01s. Panel A is for the outcome variable of whether an employee is paid by the PI of the renewed R01. Panel B is for the outcome variable of whether an employee is paid on any grants.
Table 1.
Static difference-in-differences estimates.
Fig 5.
Proportion of renewed R01s that experienced an interruption by fiscal year.
An interruption is defined as a gap in funding of more than 30 days. Source: NIH ExPorter.
Fig 6.
Variation in interruptions across NIH Institutes and Centers (ICs).
The figure shows the proportion of interrupted projects by fiscal year for the National Cancer Institute (NCI) and National Institute of Allergy and Infectious Diseases (NIAID). Source: NIH ExPorter.
Table 2.
Example of NIH ExPorter data before aggregation into project periods.
Table 3.
Full list of application types for NIH grants.
Detailed definitions available at https://grants.nih.gov/grants/how-to-apply-application-guide/prepare-to-apply-and-register/type-of-applications.htm.
Fig 7.
This is a histogram of the ratio of total negative to total positive expenditure amounts for a PI, as described in the section on negative transaction amounts in UMETRICS.
The ratio is given a value of zero if total positive expenditure was zero and total negative expenditure is also zero.
Fig 8.
This is one of two screenshots of tables from the UMETRICS 2019 manual describing the employee occupation categories.
Fig 9.
This is one of two screenshots of tables from the UMETRICS 2019 manual describing the employee occupation categories.
Fig 10.
This figure shows the distribution of the time between expiry and renewal for R01 project periods that were not interrupted, i.e., renewed within 30 days.
Each unit of observation is an R01 project period. The figure is a histogram with 1-day bins.
Fig 11.
This figure shows the timing of R01 expiry for each PI-R01 combination in the analysis samples.
Figure A shows the proportion of PI-R01s from the UMETRICS sample where expiry occurred in a given year, while Figure B shows the same for the ExPORTER sample.
Fig 12.
This figure shows event-study estimates (with 95% confidence intervals) of the difference in spending between PIs of interrupted and uninterrupted R01s.
Each panel shows the estimates for a different outcome variable: total expenditure by PI, total vendor expenditure by PI, total labor expenditure by PI, and total number of employees paid by PI. Month 0 is the month that the expiring R01 expires. Month -11 is the excluded category for the regression. Regressions are run separately on subsamples of PIs that have one R01 grant (green) or multiple R01s (brown), including R01-equivalents and P01 grants. Standard errors are clustered by expiring R01 project period.
Fig 13.
This figure shows event-study estimates (with 95% confidence intervals) of the difference in spending between PIs of interrupted and uninterrupted R01s.
Treated and control groups are matched on length of the expiring R01 project period, NIH IC, and university, and the regression is weighted using the matching weights. Each panel shows the estimates for a different outcome variable: total expenditure by PI, total vendor expenditure by PI, total labor expenditure by PI, and total number of employees paid by PI. Month 0 is the month that the focal R01 expires. Month -11 is the excluded category for the regression. Regressions are run separately on subsamples of PIs that have one R01 grant (green) or multiple R01s (brown), including R01-equivalents and P01 grants. Standard errors are clustered at the expiring R01 level.
Fig 14.
Average total direct expenditures (arcsinh transformed) per month for interrupted and uninterrupted projects, separately calculated for Principal Investigators with one R01 and those with at least two R01s.
Fig 15.
Histogram of total direct expenditures for each month relative to R01 expiry.
Unit of observation is a PI-R01 period.
Fig 16.
This graph shows event-study estimates from a balanced panel of R01-PIs 12 months before and after the focal R01’s expiry month, covering a period of 24 months.
Separate event study coefficients are estimated for interruptions that were 31 to 90 days and interruptions that were more than 90 days. The regressions include R01-PI fixed effects and relative-to-expiry month fixed effects. Month 0 is the month that the project’s budget expires. These regressions are run separately on subsamples of PIs that have one R01 grant (left) or multiple R01s (right), including R01-equivalents and P01 grants. Month -11 is the excluded category for the regression. 95% confidence intervals are clustered by expiring R01 project period.
Fig 17.
This figure shows event study estimates (with 95% confidence intervals) of the difference in spending between PIs of interrupted and uninterrupted R01s.
Estimation is done using the Callaway-Sant’anna (2020) doubly robust estimator, including the length of the expiring R01 project period as a covariate. Each panel shows the estimates for a different outcome variable: total expenditure by PI, total vendor expenditure by PI, total labor expenditure by PI, and total number of employees paid by PI. Month 0 is the month that the focal R01 expires. Month -10 is the excluded category for the regression. For comparison with the main results, the point estimates and confidence intervals have been adjusted so that the point estimate for Month -11 is 0. Regressions are run separately on subsamples of PIs that have one R01 grant (green) or multiple R01s (brown), including R01-equivalents and P01 grants. Standard errors are clustered by expiring R01 project period.
Fig 18.
This figure shows event-study estimates (with 95% confidence intervals) of the difference in spending between PIs of interrupted and uninterrupted R01s.
Each panel shows the estimates for a different outcome variable: total expenditure by PI (A), total vendor expenditure by PI (B), total labor expenditure by PI (C), and total number of employees paid by PI (D). Month 0 is the month that the focal R01 expires. Month 11 is the excluded category for the regression. Regressions are run separately on subsamples of PIs that have one R01 grant (green) or multiple R01s (brown), including R01-equivalents and P01 grants. Standard errors are clustered by expiring R01 project period.
Fig 19.
This graph shows event-study estimates from a balanced panel of R01-PIs 12 months before and after the focal R01’s renewal month, covering a period of 24 months.
Separate event study coefficients are estimated for interruptions that were 31 to 90 days and interruptions that were more than 90 days. The regressions include R01-PI fixed effects and relative-to-expiry month fixed effects. Month 0 is the month that the project’s budget was renewed. These regressions are run separately on subsamples of PIs that have one R01 grant (left) or multiple R01s (right), including R01-equivalents and P01 grants. Month 11 is the excluded category for the regression. 95% confidence intervals are clustered by expiring R01 project period.
Table 4.
Count of employees paid by PI at one year before expiry.
Fig 20.
This graph shows event-study estimates from a balanced panel of R01-PIs 12 months before and after the focal R01’s expiry month, covering a period of 24 months.
The same specification is estimated for each occupation separately, where the outcome is the total number of employees of that occupation paid by the focal lab/PI. The regressions include R01-PI fixed effects and relative-to-expiry month fixed effects. Month 0 is the month that the project’s budget expires. These regressions are run separately on subsamples of PIs that have one R01 grant (top) or multiple R01s (bottom), including R01-equivalents and P01 grants. Month -11 is the excluded category for the regression. 95% confidence intervals are clustered by expiring R01 project period. Percentage changes (plotted as text) are calculated using the median number of employees for interrupted labs at month -11 as baseline.
Fig 21.
The left column of this figure plots the average probability every month that an employee is paid by the focal PI.
The right column plots the average probability that an employee is paid by any grant at all. Employees linked to one R01 are represented in the top row. Employees linked to 2 or more R01s are represented in the bottom row.
Fig 22.
This figure plots the average probability of being paid by the same PI or any grants at all in a given month for employees on interrupted (green) and uninterrupted projects (red).
The data are also subset by employees associated with only one R01-equivalent or 2 or more R01-equivalents.
Fig 23.
This figure plots the event study coefficients estimating the difference in publication counts (arcsinh-transformed) between PIs that had an interrupted R01 and PIs that had a continuously funded R01, relative to publications in the year of R01 renewal.
R01-PI and treatment cohort-calendar year fixed effects are included. 95% confidence intervals are clustered by expiring R01 project period. The left/red plot is for PIs that only had one R01 and the right/blue plot is for PIs that had equivalent grants.
Fig 24.
This figure shows “static” difference-in-difference estimates and 95% confidence intervals of the difference in publication output if a PI had an interrupted R01, estimated separately on NIH IC subsamples.
The regression includes treatment-cohort-by-year and PI-R01-renewal fixed effects. Dependent variables are raw publication counts, arcsinh-transformed. Standard errors clustered by expiring R01 project period.
Fig 25.
This figure shows “static” difference-in-difference estimates and 95% confidence intervals of the difference in publication output if a PI had an interrupted R01, estimated separately on subsamples divided by whether the PI was above or below the median career age in the sample.
The regression includes treatment-cohort-by-year and PI-R01-renewal fixed effects. Dependent variables are raw publication counts, arcsinh-transformed. Standard errors clustered by expiring R01 project period.