What does "data pooling" actually mean?

The program Distance enables pooling of data across sites and derivation of density estimates at those sites (strata) where there are too few records to fit separate detection function.


A car pool is an arrangement where several people share use of their cars, often for regular trips to work/school/shopping. The idea being that with more people in each car, less cars are actually used in total, so it's largely a matter of efficiency.

In the more general case, we pool our resources so that collectively we make better use of them. In the computing sense, data pool can be slightly misleading, because it often just means a centralised database. Strictly speaking, it ought to mean an arrangement whereby multiple distributed data servers store "their own" data locally but provide access to that data across the entire network. In practice, it's a buzzword that's often used loosely.


Generally speaking, this is the same as "pooling your resources", i.e. combining all of your initlally-separate resources into a single large pool so as to enable larger efforts and results than anyone could have made individually.

So, 'pooling the data' means combining all of the data points from various sites into a single large collection so that you can run detection functions on the entire collection and calculate a density estimate, where if you had run the detection on a single site's data set, there would not have been enough data in the set to make a decent estimate.

  • I worked in the UK bus indistry, helping companies store and access vast quantities of historical ticket data. Most companies were part of larger groups, who often wanted to analyse multiple companies' data collectively. Sometimes individual companies wanted to compare themselves to others in the group. We didn't create centralised databases because that would have been cumbersome. We created "data pools" by setting up query facilities allowing each company to access the databases of all the other companies. This I feel more accurately reflects the usage we're talking about here. – FumbleFingers Oct 9 '11 at 15:51

from Lazar et al., 2002

In this paper we review some popular methods for combining information, and demonstrate the surveyed techniques on a sample data set.

Wong et al., 2008 cites Lazar et al, 2002:

for a review of some common data pooling methods, see Lazar et al., 2002).

given the context of how Lazar et al., 2002 states the goal:

"Our goal, then, is to explore existing statistical techniques for combining data in the functional neuroimaging context. We don’t think that there will be any argument with the goal itself; there is more power in data based on multiple subjects than there is in data based on one—this is the general statistical principle that we can learn more from a (well-chosen) large sample than from a small sample. Combining the data from many subjects will also result in a stronger signal; that is, if there is a reliable effect to be seen, pooling the data in a suitable way should make it more obvious. Finally, there is the issue of being able to generalize from one sample to a larger population. We extrapolate our inference from the specific subjects that we scanned, to others like them. As we pool together the data from more subjects, we are better able to make this generalization."

based from that statement, we can probably deduce that data pooling is the combining of data from different sources so that it will result in a 1. "stronger signal" and 2. a better generalization from the combined samples to a larger population. My educated guess is that there's a better result or better generalization from the data pooled vs. from a set of sparse data.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.