RIVETS stands for Rounded Input Variables, Exact Test Statistics.

Scientific papers report truncated data. Say we had a t-test that said (both n=30, mean=2.5, SD=0.6 vs. mean=3.2, SD=0.8, t=2.14, p=0.037). This is actually surprisingly little information, because the redacted decimals can sometimes have a big effect on the outcome of the test!

In fact, if all we know is the above truncated data, the true answer could be anywhere from t=2.54 p=0.013  (the biggest possible difference) to t=1.77, p=0.082 (the smallest possible difference).

So, what should we make of a paper when we see all the test statistics are calculated from the truncated 1dp data? We should assume that someone cut those decimal tails off and calculated them manually, rather than letting their computer do it.

Now, that's a strange decision... could there we a nefarious reason that a researcher would hand calculate statistical tests?

(Note: there certainly could be.)

Some additional fun points:

• RIVETS isn't a frontline test, but it's one of my favourites. I just think it's a cute observation. Nick doesn't like it as much, even though he came up with it.

• RIVETS also works on beta statistics from regression tables, because you can calculate the t value (b/SE = t).

• RIVETS was also developed on the Stewart papers (see DEBIT). 

• RIVETS has R code available, it's in the link above.

Go back Home