Examples of 21st century Small Area Census Research

There now is widespread enthusiasm for what I call "one off" mega-sample studies to estimate population health parameters of interest across the full range of suspected genomic and enviromic determinants of causal or protective significance (1). To define "mega-sample" I generally work from a concept of multiples of a scaled single Principal Investigator project (e.g., a five year NIH R01 project within the standard NIH budget limit of US$500,000 direct costs, minus overhead, each year). Here, the working model for this type of R01 project starts with the concept of a population being sampled and measured on a scale that is within the grasp of a small research team led by one Principal Investigator, with at least one face to face assessment session with each participant, lasting 45-90 minutes, and at least one followup assessment session roughly one year later. What can be achieved on the scale of a single R01project of this type is a local area target population sample of size roughly n=2,500, assuming 80% participation level. That is a project on a manageable scale for a single well-trained population health scientist.

 For a "mega-sample" project, think of multiples of this basic project working unit, with two important implications:

First, more than one PI will be required for management of a second project on this scale in order to maintain quality control and protect against drift of assessment protocols. This situation is due to the extended time required to achieve larger samples, as well as 'house effects' and 'sub-house effects' created by organization of field work across time or across the increased numbers of project directors and fieldwork supervisors required to shift the scale upward beyond the scale of the first project as described above.

B. There is no chance for systematic replication. The wad is shot on getting one mega-sample rather than the option of multiple studies that can shed light on reproducibility and robustness via production of multiple estimates of the same study parameter.

Yes, the mega-sample can be sorted into batches after the fact, but consider the additional waste in the form of opportunity cost-- fewer new investigators engaged in field research early in their careers because only senior investigators will be judged as meritorious leaders of the mega-sample units.

The problem is compounded when the centrally designed "U" cooperative agreement mechanism is substituted for R01 multi-PI linked investigations, which gives scientists in the funding agency too much of a controlling hand -- unless from the get-go they have the best ideas, as judged by non-government scientists. I have seen this happen only rarely.

Exceptions can occur. The NIMH ECA program began with a contract to Prof. Jerry Myers at Yale and then was extended to a five site study (e.g., Prof. Mort Kramer at Hopkins, Lee Robins at Wash U St. Louis, Prof. Dan Blazer at Duke, Prof. Marv  Karno at UCLA). Later, led by Bill Eaton at the Baltimore site, our team was able to extend that U project via competitive R01 proposals and it now is being sustained with a new award for another followup, more than36 years after we drew the Baltimore sample and took baseline assessments in 1981. But this kind of enriched longitudinal study of a local area occurred at only one of the five 'U' sites. We ought to be able to do better. .One reason, I think, was the centrally controlled 'U' origin of that study. In another universe perhaps there is a multi-PI version of linked R01 from the get-go. My guess is that more than one of the five sites lasted longer than a single award cycle for data gathering in that universe, versus the one-fifth success value  achieved with the centrally controlled 'U' starting point.

See if you can find 21st century examples of PI-initiated local area census work, of the type launched by Hopkins in Hagerstown, Maryland, and post comments about them so that we can study them. 

Here is a link to a very brief description of the Hagerstown local area census study. Please add a comment if you can find a better description in open source online document. I have been unable to find one.

See pages 110-111 of this book, which you can view using google books. It is from Whatley T et al., Am. J Epidemiol, 1968 (vol 89, no. 1).