Note the "wiggle" in the vertical axis to indicate to the viewer that it does not return to zero. Although the vertical scale might be regarded as exaggerating the effect of the crackdown (traffic fatalities drop by 12%), it is probably not seriously misleading.

The real problem with this graph is that only two points are plotted. The viewer of the graph, who knows about the 1956 crackdown on speeding, is liable to attribute the sudden drop in traffic fatalities to the crackdown. However, this impression is misleading, as the following graph shows:


This graph displays traffic fatalities from 1951 to 1959, including the drop between 1955 and 1956. On this presentation, the 1955-56 drop is far less striking. On the one hand, after the crackdown the number of traffic fatalities continues to decrease from 1956 to 1959, so perhaps the crackdown has had a prolonged effect. On the other hand, there are equally large drops from 1951 to 1952 and from 1953 to 1954, before the crackdown, which suggests that the drop after 1955 could be just due to the natural variability in the number of fatalities from year to year. What's more, the crackdown comes after (and possibly because of) a particularly high number of fatalities in 1955. After such a peak, the number of fatalities is likely to fall even without any intervention; this is the phenomenon of regression to the mean mentioned previously.

The source of the data is Donald T. Campbell, "Measuring the Effects of Social Innovations by Means of Time Series" in Judith M. Tanur et al., eds. (1989), Statistics: A Guide to the Unknown. Pacific Grove, CA: Wadsworth. This article contains an extended discussion of this example and related problems.


© 2004-2021 Joe Lau & Jonathan Chan