Well-designed graphics can guide the initial exploration of data and can show results of formal inference. However, too often we rely on “cute” graphics designed for the corporate boardroom that hide relationships and bias interpretation. Graphics can help present all the data in revealing ways for modest-size experiments. Many of today's experiments involve massive amounts of data and many questions that can productively be condensed into graphics that summarize detailed patterns and guide further investigation. This paper is based on an invited talk at the ASHS 2004 annual conference.
Bottom line: A picture is worth a thousand words. This overused phrase applies quite strongly to scientific data presentation. Graphics help us to spot unusual patterns in preliminary investigation and later to present key relationships in a concise and convincing manner.
Large tables of numbers may be important for documenting certain studies, but they are almost always better hidden from view. It is too easy for the eye to gravitate to spurious patterns and miss the main story. Such visual confusion can bias our perception and sidetrack useful research, providing misleading clues that can waste time and money. Why not organize material creatively into pictures that show relationships objectively? Graphics can even include cues to inference, such as se or lsd bars that show the extent of variation relative to mean tendency.
Creative graphics are not always useful; on the contrary, they can be quite harmful. For numerous examples in a variety of settings, see Wainer (1984) and the books of Edward Tufte (1983, 1990, 1997). The classic How to Lie with Statistics by Huff (1993) contains many examples of poor data analysis, often illustrated or “enhanced” with misleading graphics.
Excellent examples of graphics can be found in Tufte (1983, 1990, 1997) and Cleveland (1993, 1994). Both Tufte (1983) and Wainer (1984) show a famous diagram of Napoleon's defeat march across Europe, including details of geography, troop size, and weather conditions. Many of the best graphic examples are decades or centuries old. Although many are from designed experiments, others are used in everyday life, such as maps of transit systems. Cleveland (1993, 1994) focuses on more experimental settings, presenting a wide array of excellent graphical devices.
We illustrate some good and bad portrayals using plant breeding data from a Brassica napus L. cross of Tom Osborn. My contact with this work originated with Ferreira et al. (1995), with the map fully developed in Kole et al. (2002). These data are freely available as part of my R package bim (discussed later).
FerreiraM.E.SatagopanJ.YandellB.S.WilliamsP.H.OsbornT.C.1995Mapping loci controlling vernalization requirement and flowering time in Brassica napus Theor. Appl. Genet.90727732
KoleC.ThormanC.E.KarlssonB.H.PaltaJ.P.GaffneyP.YandellB.S.OsbornT.C.2002Comparative mapping of loci controlling winter survival and related traits in oilseed Brassica rapa and B. napus Mol. Breed.9201210