Despite our scientific training, when we think about ways to build a more diverse biomedical research workforce, we may base our ideas on sentiments and preconceptions rather than the best evidence. One way to avoid this is to approach the challenge of increasing diversity as a scientific problem.

The first step is to understand the scope of underrepresentation, which is discussed elsewhere in this supplement and in National Science Foundation reports. What's clear in that data is that achieving proportional representation among new PhDs in the sciences would require us to produce about 1,700 additional minority PhDs per year, and even at that rate it would take many years to achieve parity in the workforce.

National Institutes of Health undergraduate training programs at minority-serving institutions provide a total of 800 slots for juniors and seniors and should lead to 400 baccalaureate degrees per year. If every one of these students were to progress on to a science PhD, these programs would contribute significantly to diversity. But not all students given a supported research opportunity go on to a research career, and we can't assume that the NIH trainees who do go on to a PhD represent an increase above a historical baseline. Moreover, tracking individual participants, while valuable in many ways, will not tell us whether the effort increased absolute numbers.

The second step is to build upon the work of others. There is a growing literature on the barriers that minorities and women face on their career paths, and also on how and why specific interventions succeed. But much of the literature is for specialists in various fields of psychology and sociology and needs to be critically reviewed and made more accessible to scientists in other fields who are interested in contributing to change.

Next, we must identify what each program is intended to achieve. All big problems such as cancer, heart disease, and AIDS must be attacked on multiple fronts by discrete but coordinated efforts. Underrepresentation is a similarly huge, complex problem, and it makes no more sense to expect that a single intervention - improving the research infrastructure at minority-serving institutions, for example - will solve it than it would for a single intervention to solve any of these other complex problems.

Then we need to identify multiple strategies with specific aims and milestones to use in measuring progress. Our planning must include estimating the extent to which we can improve outcomes by expanding the pool of potential minority researchers, as well as identifying and being more successful in retaining those already engaged.

In designing programs, it's important to recognize any assumptions we are making. For example, we assume that a problem is solvable, that talent is not limited to any group, and the skills needed to be a productive researcher are teachable. Also, we assume that exposing students to laboratory research will inspire them to consider research careers and motivate them to improve their overall academic preparation. And we might presuppose that students are ready for this exposure at a specific stage in their education as well as that most labs are willing and able to provide students with mentored experiences. Assumptions could also be made about the level of institutional involvement or the availability of resources.

Ideally, we would determine at the outset whether our assumptions are valid. If we can't be sure, we would need to take these uncertainties into account in designing programs and acknowledge them in discussing outcomes.

Just as we publish our scientific results for others to scrutinize, evaluation and sharing of outcomes must be critical elements of our diversity strategies. Thinking of evaluation in terms of accountability purposes can undermine it as an endeavor for understanding and self-improvement. Evaluation is made more challenging by the difficulty in understanding how the context of a particular program influences its success or failure. And how do we judge the "added value" of an intervention or the relationship between cause and effect? Selection bias and other variables must always be considered.

We also need to understand the efficacy of program components. This kind of assessment goes beyond the bounds of what is commonly considered evaluation, and its complexity presents us with additional challenges. While evaluation focuses on overall results, such as whether students involved in a particular program went on to receive PhD degrees, efficacy studies might attempt to tease apart the causal relationships between specific program elements and desired behavioral changes or skill acquisition.

We already have the tools we need to devise a productive approach to achieving diversity in the biomedical research workforce. The process should look much like our research approach to other big problems. We have ideas, we experiment, we collect and analyze data, and we share the results. We seek a diversity of ideas and we encourage thoughtful engagement. Fresh perspectives and skepticism are of as much value as longstanding involvement with the issue. We expect that our work will generate new insights and lead to significant progress.

The NIH has begun to work along these lines (see sidebar). However, efforts to develop the breadth of talent in this country are too important to be isolated in select offices or targeted programs. Inclusiveness and diversity matter, greatly, and every scientific program administrator, investigator, and grantee institution should be concerned with them.

Clifton A. Poodry directs the Division of Minority Opportunities in Research at the NIH's National Institute of General Medical Sciences.