'Tis the season to submit major research proposals to NSF -- if you haven't done so already. NSF's target date for CISE proposals was November 1, but November and early December proposals are likely to compete evenly. Proposals later than that are less likely to complete review before funds are spent out.

[I laid out some of this in V1 N7. I speak for myself, not for NSF. My comments are mainly for the Division of Information, Robotics, and Intelligent Systems (IRIS) within the Directorate of Computers and Information in Science and Engineering (CISE). I think most CISE and Engineering proposals are handled similarly, but individual programs may differ. Networking and Communications has December 1 and May 1 deadlines, and NSFNet accepts proposals at any time. Biology and Linguistics use panel review, which is faster (but prone to superficial evaluation or intellectual bias if not carefully managed).]

NSF has many different programs, and many initiatives and competitions within and across programs. Each has its own statement of purpose and of review criteria, and each has unwritten policies based on expediency, custom, politics, and personal preferences. Some programs are only open to women or to researchers at minority institutions. Most competitions give very little guidance on proposal content, supposedly to reduce the hassle factor and to let you present your work in its best light. This lack of guidance is a hurdle for new investigators, and makes review more difficult.

The Research Initiation Awards (RIA) competition has a deadline two or three months from now. It's limited to tenure- track faculty members who have not previously had their own government grants. Past competitions have been limited to two years of funding at $30K per year, plus $10K in hardware. (NSF is trying to increase its average grant above two years, but I doubt that RIA will be extended to three years.) Proposals must be short and conform to a well-specified format. Evaluation is by panel, and winning is definitely easier within the competition than if these same proposals competed in the regular programs. It's also a good place to introduce new research areas for which NSF has no established program.

A few program officers shape their programs and lead the charge into the future, despite consternation from those outside the chosen thrust. DARPA favors this approach, as it's a good way to get maximum technological impact from limited funds. NSF usually goes for maximum political impact instead, which means distributing much of its money to people outside the mainstream. The heads of big labs then object, and they have considerable representation in NSF's advisory bodies. So the path of least furor is to choose good reviewers and let them bear the responsibility. (This makes Congress nervous, but in the end peer review is always seen to be necessary.)

When proposals arrive, they go to Proposal Processing for entry into the tracking system. This takes about a month. (Want faster service? Lobby for increased administrative funds. NSF gets about 5% of the funds it handles -- a third of United Way's overhead.) Proposals are then distributed to divisions and to programs. You can avoid an unfortunate assignment by designating the program you want on your cover page. (You did write for a specific review community, didn't you? NSF programs correspond roughly to major scientific journals and conferences, although new technologies seldom have their own programs.)

The program officer -- usually a Program Director (PD), but it could be an Assistant Director or someone just filling in -- scans each proposal to make sure it fits the program. Obvious misfits are sent to appropriate programs. Multidisciplinary proposals are passed around for possible buy-in by other programs, subject to good reviews in each discipline. (NSF loves multidisciplinary proposals, but it's tough to write one that appeals to more than one set of reviewers.) Borderline or vague proposals are traded or "sold around" until someone accepts responsibility -- a process that sometimes takes months. You can avoid delays if you've written for a specific program and have stressed the importance of your work in the abstract and introduction.

NSF does not have a program for every line of scientific inquiry; it just seems that way because they seldom return proposals unreviewed. Division directors (DDs) have the authority to reject proposals lacking relevant research content, but neither they nor the PDs wish to offend submitters. (Such actions can be appealed, and are also likely to spark protests to Congressmen -- especially if minorities or small schools are involved.) Even suggesting that a proposal be withdrawn can be a bit risky, so proposals are usually sent for review no matter how badly they fit program goals. If you'd rather negotiate than be killed in anonymous review, discuss content with the PD before submitting. A two-page white paper may be appropriate, but expect to call at least once before the PD will find time to read it. (I recommend that you skip the white paper and just call or send email.) Obtaining program literature can also help you to write a targeted proposal, especially if you can get the PD's own write-ups. Descriptions of special initiatives provide a good clue to the PD's current interests.

Some people, especially deans and department heads, like to drop by and chat with program officers. PDs accept this as part of the job, and some really do enjoy a chance to talk about what's going on in the field. If the program is a busy one, though, any visit of more than 15 minutes is an imposition. Bureaucrats are seldom eager to meet with the people that they serve. A short visit is fine if you're looking for guidance, but guidance can be given over the phone and literature can be sent by mail. A typical PD -- one who relates to technical proposals rather than infrastructure and politics -- has nothing to gain from your visit. Any past results you can show are irrelevant, and an outright sales pitch will be resented. (I disliked any process that might give more influence to wealthy or nearby labs than to those that couldn't afford a visit. I avoided lab visits for similar reasons.)

Videotapes are like salesmen than can't be speeded up or reasoned with.

Many principal investigators (PIs) worry that program directors will deliberately choose reviewers in order to get a predetermined outcome. An experienced PD can do this, but has no real need to. (If a proposal doesn't fit program goals, the PD can recommend declination no matter what the reviewers say.) Most PDs are inexperienced, though, and haven't the vaguest idea what reviewers are going to say. The only exception is for a resubmitted proposal. Some PDs send it to the original reviewers to see whether previously cited faults have been corrected; others (like myself) would rather get a clean slate of reviewers. (I've heard the first called "being fair to the process," the second "being fair to the PI.") There's nothing unscrupulous about resubmitting a proposal, any more than there is in direct- mail advertisers sending you repeated offers. Unfortunately, many PDs and reviewers are offended by unmodified resubmissions, so I'd recommend that you do a thorough rewrite. I don't ordinarily recommend that you point out the situation, as that can make the past [unfavorable] reviews part of the current evaluation.

There's also nothing wrong in submitting multiple proposals to NSF, as long as each is a distinct piece of work. You do have to state in the proposals that you're doing this. You can't submit the same request to multiple programs, and you can't submit proposals differing only in level of effort or budget -- which are negotiable in any case. Since funding decisions are partly stochastic, it makes great sense to submit multiple proposals. Note, however, that once you've won funding for any one of them, PDs will feel that you've had your share. (They can't pay you twice for the same summer months, of course.) If you give them a choice between a full $70K/year proposal and a $15K participation in another, they're very likely to take the cheaper way out. PDs may also take each proposal less seriously if you have others, but I think there is a tendency to ensure that at least one is funded.

Reviewer assignment is a "Dutch marriage" problem, the matching of a pool of reviewers to a pool of proposals. No, you don't necessarily get the obviously best reviewers; they may be needed for other proposals. PDs assign the best reviewers they can come up with, but time and expertise may be limited. To improve the selection, suggest five or ten reviewers in your cover letter. NSF is now encouraging this for all programs. You can also specify individuals who should not be used as reviewers, either by name or by quoting their comments in past reviews. Reviewer selection is up to the PD, but ignoring your suggestions would require written justification. (Unfortunately, instructions in your cover letter may be overlooked at reviewer assignment time. I recommend attaching a yellow sticky note to your proposal to flag any special handling.) If you make no suggestions, you are likely to get anyone mentioned in your bibliography who is also in the NSF database. If you've only referenced your own work, the PD may search journals and conference proceedings for possible reviewers or may call you for suggestions.

I should mention one caveat. Any reviewers you suggest will be presumed knowledgeable and possibly biased. Criticisms from them will be given extra weight, whereas praise may sometimes be discounted. I believe that this is a small price to pay to get reviewers who really are knowledgeable (and perhaps even biased!). Besides, you'll write a better proposal if you know your audience.

Five to eight reviewers are selected for each proposal. (I've seen as many as 18 for multidisciplinary proposals.) Some will be chosen for technical expertise, others for breadth of experience. Some may be known supporters of your approach, others known critics. A few may be graduate students who have recently published in your area. (This is excellent training for the students, and a necessary entrainment if the reviewer base is to grow. Besides, graduate students often give the longest, most detailed reviews -- and sometimes the most insightful ones.) A good PD will seek a broad range of opinions, but will keep their sources in mind when making a final evaluation.

PDs seldom contact reviewers in advance, as they just don't have time. Mailing the proposal out takes a week or so after reviewers are assigned. Prompting letters are sent if three reviews are not received in six to eight weeks (or sometimes much longer, if the PD just can't get to it). If prompts don't work, the PD can call reluctant reviewers or send the proposal to additional people. (Assigning new reviewers adds months to the process, but takes the least time for the PD. New reviewers can be selected from scientists who were busy on the first pass, so this often works out well.) Sometimes the second-round reviewers demure, and a third round is needed. Such complications occasionally delay a proposal for 18 months or more, although NSF is targeting nine months for total processing.

Eventually the reviews are in and the PD finds time to write detailed recommendations. A PD only recommends action to the DD, who recommends award or declination to NSF's Grants and Contracts. In most cases, though, it is the PD's judgment that prevails. The PD's judgment generally matches recommendations of the peer reviewers, but reviewers do NOT make the decision. (In panel reviews, it is common to use panelist evaluations to rank order proposals; NSF has final responsibility, though, and can second- guess the cutoff levels and the rank ordering.)

Obvious declinations are processed first -- although NSF is applying pressure for FIFO processing when possible. Obvious awards are also made, and such early awards take only a month or so to get through Grants and Contracts. Marginal proposals are often held as long as possible in order to evaluate their competition. If you need summer funding, be sure to make it plain in the proposal and in a later call or letter to the PD. Your institution will usually let you spend funds once your proposal is recommended to Grants and Contracts.

If you're really strapped for time, request expedited processing. NSF can compress six weeks' processing into six hours if necessary, especially if an NSF error created the need. In other cases, consider submitting a proposal under the Small Grants for Exploratory Research (SGER) program. These are urgent requests for small amounts of support, to be awarded or declined in about three months. (SGERs can also be used for innovative research where no review community exists.) Each program can spend up to 5% of its funds on such projects, almost entirely at the discretion of the PD.

A decision for award or declination may hinge on intuition, but the written justification should echo specific concerns of the reviewers -- tempered by the understanding of the PD. A PD typically looks for a consensus attitude within the reviews. Reviewers tend to argue for or against funding, and this presence or absence of enthusiasm is at least as important as any summary check box on the form. PDs also factor in program goals, their own evaluations, available budget, effect on scientific infrastructure, legally or politically mandated diversities, and any announced program-specific factors. Still, I'd guess that reviewer enthusiasm is the number-one factor.

The number-two factor may be "political potential," the PD's ability to sell the work to others -- especially to Congress. You increase your chances of winning if your abstract and introduction make that sales job easy. From another viewpoint, PDs are always looking for a sexy new problem or an important breakthrough. Each wants to do some good with his or her discretionary funds. Your proposal should offer something the PD can be proud of supporting. If you can't explain the importance of your work over the phone, or in a 30-second "elevator summary," you probably can't write a good proposal.

A third factor, although perhaps very minor, is "negotiability." Some proposals are set up as a choice: give me full funding and I'll do A, B, and C; give me less and I'll only do A. This, or negotiable budget items, make it easy for the PD to call up and discuss cost-saving options with you. Once you've gotten that far, it would be difficult for the PD to recommend declination. (If funds run out, the PD can hold your proposal for funding with next year's money. Most hate to do it, but they may -- once they've admitted that your work is of interest. You'll also be first in line for any funds that show up at the end of the fiscal year.)

Keep in mind that this is a competition. Continuing grants get the first cut, then special programs get theirs, and regular proposals must divide up the rest. With current proposal pressure, few NSF programs can make awards to less than excellent proposals, as judged by perhaps a third of the reviewers. NSF has money for about a fifth to a quarter of the regular proposals -- and not the full amount requested, either. My first year, I was asked for $80M; I only had $5M -- and less than $2M was available for new grants. PDs necessarily look for reasons to turn down proposals, especially expensive ones.

Declination by NSF is not a rejection of your work. Think about the junk mail you get every day. Declining an offer doesn't mean you denigrate the product; you simply have other uses for the money. NSF certainly wants to fund the best proposals, but the evaluation process is stochastic. When dozens of proposals get equivalent ratings, it makes little difference to the American people whether one good proposal or another is funded. In such company, the funding decision may depend more on your proposal- writing skills than on the inherent merit of your work.

This is important enough that I'll say it again. NSF is not a science court. Reviewer comments on the scientific quality of your work are important but seldom decisive. A PD will not understand your proposal as well as you do, and probably not as well as the reviewers do. (I, for instance, could not have been an authority on all of robotics, vision, speech recognition, pattern recognition, and AI. A temporary program officer may do just fine with very little technical knowledge.) The PD seldom has enough insight to judge scientific disputes, and is likely to fund both horses rather than risk backing the wrong one. If you anticipate technical disagreements, say so and claim that your viewpoint deserves to be explored. Proposals are likely to fail because they aren't persuasive enough, rather than for any technical reason. If you've gotten your reviewers excited, you'll win funding; if not, you won't. Whether you make the sale depends more on the problem you've chosen and the way you pitch it than on your technical competence.

Among proposals recommended for award, there's still a problem of dividing up available funds. One rule of thumb is to squeeze every PI as hard as possible. NSF tends to negotiate from a take-it-or-leave-it stance, so your skill in writing the proposal is the biggest factor in what you can get. Don't be afraid to negotiate, though; it's surprising how easily an inexperienced PD can be influenced. Negotiations are based on needs and expectations, and both are in your court. Back down on either one and you will get less money. Awards also depend on perceived potential of the work, and perceptions can still be influenced at this stage.

If you do get word from a program officer that your proposal is being recommended, you can begin spending as soon as your own administration permits it. NSF will allow you to recover costs incurred up to 90 days prior to the official award date. It takes NSF's Grants and Contracts one to two months to make an award, or even three months at the end of the fiscal year. If you do spend early, it's at your own risk. If the award doesn't come through for any reason (e.g., a government freeze), you or your institution must suffer the loss.