AN ENDLESS stream of new discoveries makes science thrilling. But, as any seasoned researcher knows, such novelties are worthless unless they can be replicated. Often, though, replication does not get done as thoroughly as it should be—or even at all. For, as any seasoned researcher also knows, replication is rarely the stuff careers are built on; studies conducted with that goal may even struggle to get published in peer-reviewed journals. In this context, a recent attempted replication is important, for it actually was published last week in a journal called Psychological Science. Its author was Michael Dufner of the University of Leipzig, in Germany. In it, he said that he was unable to replicate a fascinating previous finding which had suggested that people who smile more intensely tend to live longer than those who did not. http://newssum.netThe original study, published in 2010 by Ernest Abel and Michael Kruger, then of Wayne State University, in Detroit, seemed sound enough. The two researchers had gathered the mugshots of 196 players from the Baseball Register of 1952, which lists the game’s professionals in a given year, and asked a team of volunteers who were blind to the purpose of the experiment to study the players’ smiles and rate their intensity. Dr Abel and Mr Kruger then delved into files that contained information on when the players in their experiment had died—or if, by rare chance, they were still alive. This analysis revealed that players with full smiles were more likely to have lived to a ripe old age than those who had partial smiles or no smiles at all. Statistically, this was a strong result. The probability that it was the result of chance was one in 50. It was also an intriguing result. It dovetailed well with ideas then emerging that happiness induces biological effects which lead to improved health. Indeed, Dr Dufner often discussed the research with others and, as questions about whether the work had ever been replicated came with increasing frequency, he decided to roll up his sleeves and do so in his own laboratory. Together with a team of colleagues, he worked with a sample based on the one used in the original study. All of the players in that were included, along with a larger, non-overlapping set of 527 Baseball-Register images of players who were active slightly before, or after, 1952. Just like the researchers in the first experiment, Dr Dufner relied on “blind” volunteers to rate the intensity of the smiles in the images. Of the pictures he worked with, 40% showed no smile, 42% showed a partial smile and 18% showed a full smile. When Dr Dufner compared these numbers with how long each of the players lived, however, he found no correlation between lifespan and smile intensity in either the original or the expanded samples of players. When replications fail in this way, particularly in fields like psychology, critics often argue that “hidden moderators” like differences in time, culture or sample composition between the original study and the replicate are the reason for the failure. What is notable about Dr Dufner’s work is that it did not contain any such factors. The photos were the same. Only the volunteer examiners were different. Somehow, it seemed, the two sets had coded the players’ smiles differently and in the earlier study the chance result had come to pass, as one time in 50 it must. To check in detail what had happened Dr Dufner contacted Dr Abel, the paper’s senior author, and asked if he could see the original data. Unfortunately, these were unavailable. As Dr Abel explained to The Economist, he has now retired and when he did so he “threw out my data for the past 40 years. Time to move on.” A cautionary tale, then, of the importance both of replication and of keeping the data that a study is based on safe and sound, just in case they need to be checked again.