Selecting R&D Projects: An Informational View
R1 is a 1991 formal economics model set out in outline. Everything it concludes is a conditional result of that model rather than an empirical finding — and one of those results, a rule for ranking two projects against each other, is usable without any of the mathematics.
What R1 is, and what its propositions can support
R1 is a short academic economics paper from May 1991, published in a conference-proceedings issue. Papers of that kind are typically invited, deliberately short, and not put through full refereeing. It runs to five printed pages, sets out a formal microeconomic model in outline, and argues for a research programme rather than reporting results.
That matters. The paper's conclusions are conditional on the assumptions it states; they are not measurements of anything. Read them as if the world works like this, the following follows — a legitimate and useful form of argument, but a different one from evidence.
WHAT THE PAPER DOES AND DOES NOT CONTAIN
| Element | Present in R1? | Consequence for use |
|---|---|---|
| Formal definitions and notation | Yes — the bulk of the paper | A vocabulary for describing projects by what they learn |
| Value equations, static and dynamic | Yes — four of them | Usable as structure; no calibrated parameters |
| Tables, figures, data displays | None at all | Any visual used to teach this is your construction, not the paper's |
| Empirical numbers of any kind | None | Attribute no rule of thumb, percentage or benchmark to this paper |
| Complete proofs | No — deferred to longer prior work | Not self-contained as a proof source |
If you are assessing what a short proceedings paper can and cannot evidence, the general technique is set out in Critiquing a journal article.
Three primitives: information, project, success
The paper defines its three basic objects before introducing any notation, and the definitions are the part a practitioner can use directly.
The definitions as given
- Information
- A generalised partition of the states of the world. It is an ex ante concept — the ability to know that the true state lies in one of a set of subsets, not the ex post fact that it happens to lie in a particular one.
- R&D project
- An attempt to obtain access to a particular information structure. Specifying a project and its intensity yields a success rate: a probability of success in the static case, an arrival rate for success in the dynamic case.
- Success
- That the information structure is acquired and can be used to condition future economic decisions. Both limbs are required.
The ex ante framing is the load-bearing move. What you buy when you fund a project is not an outcome; it is a capability to distinguish between states of the world you currently cannot tell apart. Whether that capability is worth anything depends on what decisions it lets you condition.
The paper illustrates its own definitions with academic research. You start from a specific open question — say, whether a class of models has an equilibrium. That question is the information sought. Three quite different outcomes all count as success: a counterexample, a proof that equilibria exist, or additional conditions under which existence can be demonstrated. Non-success is being unable to reach any clear conclusion.
Two consequences follow, and both transfer to industrial R&D. A project is determined by two choices — selection of the question, and an intensity corresponding to the difficulty of the methods of attack examined. And success is separable from realisation: when a project succeeds you stop working on it, but you may or may not go on to publish, commercialise, or do anything at all with the result.
A project is a pair, not a budget line
Formally, a research project is the ordered pair of an information structure and a non-negative number representing intensity. Intensity is defined as a summary measure of the research methodology in terms of its corresponding costs — so cost enters the description of the project, but it does not constitute the project.
THE PROJECT OBJECT, AND WHAT IT REPLACES
| Component | What it is | What management usually records instead |
|---|---|---|
| The information structure | The question you are trying to become able to answer | A deliverable, a technology name, or a business case |
| The intensity | How hard you are attacking the question, measured through its costs | A budget figure, treated as the project's identity rather than one attribute of it |
| The success rate | Probability of success, or arrival rate for success, determined entirely by the pair above | An unstated assumption, or one confidence percentage for the whole programme |
Formal objects of the model in R1. The right-hand column is our characterisation of ordinary practice, not the paper's.
Success rates are completely specified by the project — the information structure and its intensity. Nothing else determines them. That is a strong assumption, and the paper adds another: all agents know the success rates of all projects.
Why one R&D number is the wrong measure
The aggregation critique is the paper's most quotable point and its most usable one. If two projects acquire different information, adding their costs together produces a figure that describes neither. Two portfolios with identical spend can be buying entirely different knowledge, at different success rates, with different consequences for what the organisation can decide afterwards.
- An aggregate spend figure cannot tell you what questions the organisation is trying to answer.
- It cannot show complementarity — two projects whose combined information is worth more than the sum of the parts.
- It cannot show substitution — two projects buying the same knowledge, where the second adds little.
- It cannot distinguish a portfolio of many distinct questions from one attacking a single question many ways.
- It gives no basis for saying whether the amount is right, because the comparison it invites is not well posed once projects differ.
The dominance rule, and the free disposal of information
The single most portable thing in the paper is a rule for comparing two projects. It is stated in prose in the introduction and proved as a monotonicity result later.
Comparing two candidate projects
The value of a project in the model is the amount by which its expected rewards exceed its cost. In the dynamic version costs are paid in every period up to and including the one in which success arrives, and future rewards are discounted. The structure is familiar from discounted cash flow — see The financial frame for R&D management — but the object being valued is the acquisition of a question's answer rather than a stream of product revenue.
The firm's choice problem, and why it is not well behaved
The firm is assumed either to be risk neutral, maximising expected profits, or risk averse with a utility function over profits that is continuous, increasing and concave. Under either assumption, the optimal choice of a single project from a compact set of candidates is well defined — but need not be unique. Maximisers exist; ties may occur.
Moving from one project to a portfolio takes a specific construction, and it is worth following because it is what makes the problem tractable at all.
How a portfolio is built in the model
Expand the choice set
Move from choosing single projects to choosing finite and bounded sets of projects. The motivating image the paper offers is competing research teams inside one firm, which may be complements or substitutes for one another.
Represent each project as a point mass
Each project is represented by the probability measure that assigns mass one to it.
Sum them
A profile or portfolio of projects is a finite sum of those probability measures.
Collect the consequence
This yields a compact choice set, so the maximum theorem applies and an optimum exists.
The same non-convexity bites in competition. Modelling rival firms as playing a non-cooperative game in which strategies are sets of projects, the paper's negative result is that the game does not generally have an equilibrium in pure strategies, even when firms are restricted to finitely many alternative projects. The remedy is mixed strategies, which restores convexity of strategy sets and best-reply correspondences; the interpretation the paper prefers is each firm optimising against its probabilistic beliefs about the others' behaviour.
The usable residue is negative but real: reasoning of the form our competitor will rationally choose project X, so we should choose Y has no foundation here. Rival project choice is properly a distribution over possibilities — one more input to Sources of uncertainty in R&D projects.
Welfare consequences: a framing critique and four propositions
The paper's final section applies the informational view to public-policy questions about R&D. It opens with a critique of how the question is normally framed, then offers four propositions. Each is conditional, and the condition is the interesting part.
THE WELFARE ARGUMENT — EACH RESULT WITH THE CONDITION IT DEPENDS ON
| Result | Holds under | What it displaces |
|---|---|---|
| The framing critique: the under- or over-investment comparison is not well posed | Whenever projects acquire different information from one another | Comparing actual aggregate R&D investment against a socially optimal aggregate |
| Patent races tend to create excessive waste | If every project yields exactly the same innovation when it succeeds | The unconditional claim that racing is wasteful however projects differ |
| Duplication of effort is unlikely to be socially suboptimal | If firms pursue distinct projects, or if intermediate discoveries can be shared | The presumption that two organisations on one problem is prima facie waste |
| Spillovers cannot be modelled without differentiated project types | Where value arises by joining information from a project aimed at one purpose with another firm's project aimed at another | Spillover measures built on aggregate spend |
| Cooperative ventures should be assessed on economies of scope and specialised expertise | Where those are the actual source of the gain | Assessment by combined R&D budget or by counting participants |
All five are conditional results of the model stated in R1. None is an empirical finding; the paper reports no data of any kind.
The duplication proposition is the one most likely to change a decision inside an organisation rather than a government. Running two teams at one problem is usually treated as obvious inefficiency. On this account it is clearly wasteful only when both would produce the identical innovation — if they attack the question by genuinely different routes, or partial results can be shared, the presumption does not hold.
Using an informational view without the mathematics
None of the formal apparatus is needed to change how a project register is written. The model's objects translate into questions you can ask of any candidate project.
Six questions the informational view puts on the table
- What question would this project make us able to answer? Write the question, not the deliverable.
- What future decisions would that answer let us condition? If none, the project has no value in this frame even if it succeeds.
- At what intensity are we attacking it — and is intensity recorded separately from the question, or has the budget become the project's identity?
- What is the success rate at that intensity, and does anyone outside the team share our estimate of it?
- Do any two projects buy the same knowledge? If so, the second is a substitute and the duplication argument does not protect it.
What to carry forward
- Describe a project by two things — the question it would answer and the intensity at which you are attacking it. Cost is an attribute of the second, not the identity of the project.
- Define success as acquiring knowledge you can act on, not as a discovery event or a commercial result.
- Use the dominance rule to rank pairs, and respect its conjunction: richer information at a worse success rate is not dominant.
- Refuse to report a research portfolio as one spend figure. Under this model that aggregation is inherently misleading, not approximately right.
- Treat every proposition in R1 as conditional on a stated model. It contains no empirical content, and nothing in it should be quoted as a finding about how R&D actually behaves.
Frequently asked questions
Is R1 saying that R&D budgets are useless?
It is saying that a single aggregate figure cannot support the conclusions people draw from it, because projects acquire different information and the aggregate hides both complementarity and substitution between them. Budgets remain a constraint you have to manage. What the paper rejects is using aggregate spend as the measure of what R&D is doing, or as the basis for judging whether the amount is right.
Can I use the dominance rule on real projects?
Yes, and it is the most directly usable thing in the paper. For any two candidates, ask whether one would produce everything the other would and more, and whether its success rate is at least as good. If both hold, it dominates. If information improves but the success rate falls, the rule deliberately gives you no answer and you have to value the two projects properly.
Why does the paper say duplication might not be wasteful?
Because the standard waste conclusion depends on an assumption that every successful project yields the identical innovation. Where firms pursue genuinely distinct projects, or where intermediate discoveries can be shared, that assumption fails and duplication is not presumptively suboptimal. This is a conditional result of the model, not a measured effect.
What does it mean that concavity cannot hold?
The set of possible research questions in this model is not a convex set, and value therefore cannot be a concave function over it. Practically, it means you should not assume that R&D portfolio choice behaves like a well-mannered optimisation problem with smooth trade-offs and marginal reasoning at the edges. The paper presents this as a structural fact about the objects, not a modelling choice.
How much weight should a practitioner put on a 1991 proceedings paper?
Weight it as a source of structure and vocabulary, not of evidence. It is short, invited, presented in outline, defers its proofs to longer work, and contains no data. Its value is that it names a distinction — projects differ in what they learn, not only in cost and risk — that most selection frameworks leave implicit.
Does this replace financial evaluation of R&D projects?
No. The paper's value equations are recognisably discounted expected value, so it sits alongside financial evaluation rather than against it. Its argument is about what you are valuing: the acquisition of an answer that lets you condition later decisions, rather than a predetermined product outcome.
References and source attribution
- R1 - choosing R&D projects, an informational approach. Academic economics paper presented in a conference-proceedings issue of a general economics journal, May 1991; 5 printed pages; a formal microeconomic model set out in outline with five numbered footnotes and no tables, figures or data.
- Eleven copyrighted journal articles on R&D project management, supplied as a reading set for a literature review and profiled for this library. Front matter, abstracts, framework sections, tables and figures were read; article bodies were not reproduced, and all content here is paraphrase.
- Prior literature as characterised within R1: portfolio-choice treatments differentiating projects by risk profile alone, multi-stage innovation race models, location models of research choice, parallel-research models with an endogenous number of identical projects, and knowledge-efficiency treatments with inter-project complementarity. Characterised as R1 characterises them; not independently consulted.
- Supplied teaching source for this library (research methods and research process materials). Used here for page conventions and voice only; it does not treat R&D project management.
Suggested questions for Ask KEVOS
- Rewrite my project register so each project is described by the question it answers and its intensity.
- Apply the dominance rule to two R&D projects I describe and tell me whether either dominates.
- Where does the informational view disagree with a discounted cash flow view of R&D selection?
- What can and cannot be concluded from a short conference-proceedings paper presenting a model in outline?
- How would I tell whether two projects in my portfolio are buying the same knowledge?