Overview

This chapter explains how the PIPE study turns grant amounts into a 0–100 relative intensity scale and estimates a third-degree polynomial dose–response. The fitted patent response is inverted U-shaped: modeled additionality rises from 0.059 patents at the 10th percentile to 0.640 at the median, peaks at 0.683 around the 59th percentile, then falls to 0.515 at the 75th. The peak is a location in PIPE’s observed distribution, not 59% of a budget or a universal dollar target. It describes patent additionality after three years; the absorptive-capacity account is a plausible interpretation, not a separately identified causal chain.

Learning outcomes

The 0–100 dose scale preserves PIPE award ordering; it is not a universal dollar or budget percentage.
For the main patent outcome, modeled additionality rises from 0.059 at the 10th percentile to 0.683 near the 59th, then declines.
An inverted-U curve describes changing marginal patent response, not proof that high grants have no value or that capacity caused every position.

Keywords

PIPEDose–responsePatentsIntensityAdditionalityPercentiles

The Sweet Spot: Reading the Dose–Response Curve

The PIPE evidence does more than show that funded applicants patent more than comparable non-funded applicants. It asks how the estimated response changes as the grant dose changes. That is the focus here: to read the study’s normalized intensity scale, understand why its modeled curve rises and then falls, and use the reported peak as a context-specific estimate rather than as a portable instruction to make every grant larger or smaller.

From a yes-or-no treatment to a measured dose

An average treatment effect answers a broad question: what was the average difference between receiving a PIPE grant and not receiving one, under the study’s comparison design? It does not reveal whether the first increments of funding have the same value as later increments. If one recipient receives only enough to begin a project and another receives enough to assemble a larger research effort, treating both simply as “funded” hides a potentially important relationship.

The study addresses that relationship by treating the grant amount as treatment intensity. To compare amounts that may be denominated in dollars but span a skewed distribution, it rescales grant levels from 0 to 100. A dose of 0 is the low end of the observed intensity scale, while 100 represents the highest observed grant in the estimation frame. The scale is therefore relative to the awards present in the PIPE data. It is not a percentage of a firm’s budget, a percentage of the government’s total R&D spending, or a universal dollar scale that can be carried unchanged to another program.

That normalization makes the curve easier to read, but it does not make the underlying grants interchangeable. A move from 20 to 40 on the scale represents movement through the observed PIPE distribution, not necessarily the same dollar increase as a move from 60 to 80. The distribution is right-skewed, with many awards toward the lower end and a smaller number of much larger awards. A normalized position preserves the ordering of intensity while leaving the program’s own award distribution as part of the context.

The authors estimate a dose–response function using a third-degree polynomial. In plain language, the model allows the relationship to bend: it does not force every additional unit of support to produce the same increment in patenting. The fitted function traces the expected innovation outcome associated with different positions along the observed dose scale, conditional on the study’s comparison framework. Its purpose is to expose changing marginal effects that an average recipient estimate would compress into one number. A smooth fitted function is useful because it summarizes the pattern across a continuous range rather than forcing the analysis into a few arbitrary grant categories. But smoothness is a modeling choice, not a guarantee that the real-world response changes smoothly from one dollar to the next. The curve should be read as an estimate of the relationship supported by the observed data, with the shape and location of its maximum tied to the specification and sample.

It also helps to distinguish three kinds of statement. The study observes grant amounts and later patent outcomes. The model estimates how those outcomes vary with positions on the grant scale. The capacity account interprets why the variation might have that shape. Moving from the first statement to the second requires statistical assumptions; moving from the second to the third requires a mechanism that the study does not directly identify.

Published by DeepMox

FAQ

What does 0–100 mean in the PIPE dose analysis?

It ranks grants from the low to highest observed intensity in the estimation frame; it does not represent a percentage of a firm’s budget or the government’s R&D budget.

How do the PIPE patent estimates change across the dose distribution?

They rise from 0.059 at the 10th percentile to 0.272 at the 25th, 0.640 at the median, peak at 0.683 near 59%, and fall to 0.515 at the 75th. These are modeled values, not separate experiments or award recommendations.

Does an inverted-U PIPE curve prove that high grants are harmful?

No. It shows lower modeled marginal patent additionality at higher doses within the observed setting; high grants may still produce later benefits or value outside the patent measure.

Previous9 minPIPE Counterfactual and Patent Effects

Next9 minPIPE Absorptive Capacity and Robustness

PIPE Dose–Response Curve

The Sweet Spot: Reading the Dose–Response Curve

From a yes-or-no treatment to a measured dose

FAQ