THE once-every-decade national census is a lubricant to the American political system. It gives legislatures, bureaus, and agencies the statistical flow to set representation and apportion more than $100 billion in aid from the federal government.
But almost since the beginning, there's been a stickiness to that lubrication. It has been hard - and it's getting harder - to arrive at anything like an accurate count of all the people in this growing and changing land.
The Census Bureau's solution to this perennial problem for the year 2000 is a plan for statistical sampling. It will obviate the need to deploy legions of enumerators, repeatedly and at great cost, in an effort to reach every last household. When the bureau has rounded up responses from 90 percent of the households - using both mailed forms and an initial foray of census takers - the remaining 10 percent will be filled in by sampling.
About 1 in 10 of the uncontacted Americans will be reached, one way or the other, and information about those still eluding the census will be projected from that sample. Voila! Something approaching a complete ''count.'' This technique is firmly grounded in statistical science, but it could be politically shaky. It might raise hackles among those who've felt slighted by past counts, or among those who think a constitutionally mandated ''actual enumeration'' should be literally that.
In fact, a count that makes use of sampling will probably be closer to a reliable enumeration that most past ones. It will also be more affordable at a time when Congress has already shown itself stingy in providing the Census Bureau with funds to gear up for 2000. The last count cost $2.6 billion, and the next one, using the same enumerating methods, would approach $5 billion.
Of course, the more citizens who take it on themselves to mail forms back in, the lower the cost to the government. Last time, only 63 percent cooperated.