Saturday, March 19, 2011

A pseudo issue?

Which experiment is best? An experiment were mowing is tested on 20 plots with the size of 1 square meter, or an experiment were 2 fields of 1 ha each are mown and have 50 plots each. From a restoration point of view the small plot case is lacking a lot of relevance to restoration, since the area is small and you will have edge effects. On the other hand, this is the case were you have real replicates. Many restoration papers go for the second approach were the situation is more comparable to real restoration. However, in a review the likely comment of “pseudo replication” is sure to come, even if the reviewer himself/herself thinks the chosen experimental layout is the best. The term pseudo replication was popularized by Hurlbert 1984, in a paper were he discussed the massive use of pseudo replication in published papers.

In a paper were he discussed the abuse of pseudo replication accusations Hoksanen used the term pseudo issue about automatic use of the pseudo replication argument (Hoksanen 2001).

Pseudo replication means that the samples are repeated, but the treatments are not. They can sometimes be repeated in the number of 2 or 3 but it is to few to make the detection of significant results impossible. Most restoration papers will therefore take a lot of samples in a large field were the restoration treatment is repeated once or twice, and claim that they have for example 100 replicates. And this is indeed pseudoreplication.

There is no easy way out of this problem, and perhaps it serves to highlight the use of meta-analysis were the outcome of many cases are analyzed. Unfortunately then there is the problem of bias towards positive results among published papers.

No comments: