USGS Home
Contact USGS
Search USGS

Replication requires that a sample consist of more than 1 member of a population, or that treatments be applied independently to more than 1unit. Replication provides 2benefits. First, it reduces error because an average of independent errors tends to be smaller than a single error. Replication serves to ensure against making a decision based on a single, possibly unusual, outcome of a treatment or measurement of a unit. Second, because we have several estimates of the same effect, we can estimate the error, as the variation in those estimates reflects error. We then can determine whether the value of the treated units are unusually different from those of the untreated units. The validity of that estimate of error depends on the experimental units having been drawn randomly; thus, the validity is a joint property of randomization and replication.

Is Replication Always Necessary?

Imagine yourself cooking a stew. You want to see if it needs salt. You dip a teaspoon into the kettle and take a taste. If it's not salty enough, you add more salt. Notice that you did not take replicate samples. Only one. (Further, you probably didn't randomly select where in the kettle to sample; you most certainly took it from the surface and most likely near the center of the kettle.) Cooks have been using this sampling approach for probably centuries, without evident problem. Why?

The single, nonrandomly selected sample generally suffices because the stew is fairly homogeneous with respect to salt. A teaspoon from 1 location will be about as salty as a teaspoon from another. This is because the stew has been stirred. Note that the same approach would not work for sampling meat, which is distributed less uniformly throughout a stew. Replication may not be necessary if all the members of the universe are identical, or nearly enough so.