Statistical Guidelines for Psychosomatic Medicine

Abstract

During the past two years, Psychosomatic Medicine joined many of the leading medical journals in adopting the CONSORT, MOOSE, QUOROM, and STARD guidelines for reporting clinical trials, meta-analyses, and diagnostic studies. This year, we are implementing our own set of statistical analysis guidelines. In doing so, we are joining a much smaller number of journals. Whether we are setting a trend, or simply going out on a limb, remains to be seen. We decided to implement these new guidelines for several reasons. First, they will advance the journal’s mission of promoting methodologically rigorous psychosomatic research. Second, they will enable our contributors to avoid some of the most common statistical criticisms when they submit manuscripts for review, and they will spare our reviewers from having to consider these problems. Third, they will reduce uncertainty about the acceptability of certain statistical methods. The new guidelines address several basic statistical problems that frequently appear among the hundreds of manuscripts that are submitted to Psychosomatic Medicine every year. Some of these problems result from simple failures to follow well-established statistical practices, or from adherence to statistical practices that were once widely accepted but that have since become obsolete. Others reflect legitimate differences of opinion among statisticians, or decades-long methodological controversies that have sewn persistent uncertainty and confusion among the rest of us. Still others stem from the multidisciplinary nature of psychosomatic research: Certain statistical practices are extolled in some of our constituent disciplines but are anathema in others. We are starting out with a very small set of guidelines concerning 1) directional (one-tailed) hypothesis tests; 2) artificial categorization of continuous variables; 3) automated selection (eg, stepwise selection) regression models; and 4) covariate adjustment. We will implement additional guidelines only as the need arises, and we expect the list to grow rather slowly. Although a number of newer, more advanced statistical methods such as growth curve models have begun to supercede their older, less sophisticated analytical ancestors, we are not yet establishing any guidelines favoring the former over the latter. We may do so in the future, but this is currently a low priority. We are also in the process of developing guidelines regarding testing interactions, subgroup analysis of clinical trials, and the rather complex issue of multiple testing. It is important to note that we are implementing statistical guidelines, not requirements. Although the guidelines discourage the use of certain methods, they do not prohibit them under all circumstances. For example, categorizing a continuous variable into several groups may be a reasonable approach if the functional form of the relation under study is not linear and difficult to capture using a mathematical function such as the logarithm or exponent (though there are still preferred alternatives, such as modern curve fitting algorithms). The burden will be on the contributor to make a convincing case for deviations from the guidelines. Our guidelines primarily address the choice, use, and description of statistical methods, rather than the presentation of statistical results. Our Instructions to Authors includes only a limited set of specifications for tables and illustrations. Furthermore, the CONSORT and other statements listed above comprise general reporting guidelines for specific types of articles, but they provide little guidance as to the best ways to report statistical results. However, our reviewers frequently raise questions or concerns about the presentation of statistical results in tables, graphs, or text. Consequently, we are also adopting the statistical reporting recommendations presented in two standard manuals, How to Report Statistics in Medicine (1) and the Publication Manual of the American Psychological Association, 5th edition (2). In addition, we also recommend referring to Cleveland’s 1994 book, The Elements of Graphing Data (3) as a general guide to the effective graphical presentation of data. Contributors are asked to consult these manuals as needed when deciding how to present their results, and to cite supporting material from these sources in their response if they disagree with a reviewer’s position on a statistical reporting issue. Since these manuals offer little guidance on how to report the results of some of the newer, more advanced methods, it may be necessary to consult other authoritative sources as well. Our statistical guidelines are described in a document linked to the Instruction to Authors page on our journal’s web site, www.psychosomaticmedicine.org and on our manuscript management web site, psymed.editorialmanager.com. We welcome comments on the initial guidelines and suggestions for additional ones.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 20 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.