Tipping Trends Over Time

Academic research on restaurant tipping in the U.S. that involved real tipping situations (not hypothetical scenarios or surveys of general attitudes and behaviors) was identified from personal knowledge of the tipping literature together with a Google Scholar search of “restaurant tipping.” Information about the average percent tip, as well as year and state of data collection, was extracted from those papers or from the data sets behind those papers. A total of 64 observations were obtained from 56 articles, theses, and dissertations. More details about the variables recorded are presented below and the complete data set (including source citations) is available upon request from the author. Percent tip. In the vast majority of cases, average percent tip was calculated from individual transaction-level tip and bill amounts. However, in rare cases it was calculated from study-level (Cho, 2014: Lynn and Graves, 1996:S1) or server-level (Jiang and Galm, 2014; Lynn and Simons, 1998; Pierce, Snow and McAfee, 2015) mean tip and bill/sales amounts. In one case it was calculated by averaging state level median percent tips (Lynn, 2022). Means included tips of zero for most studies, but were based only on non-zero tips for Flynn and Greenberg (2012), Lynn (2023), McCrohand and Pearl (1991), Olson (2017), and Susskind and Curry (2016). Fortunately, stiffing is U.S. restaurants is rare, so inclusion versus exclusion of tips of zero is unlikely to strongly bias the resulting means. Articles/papers with more than one study, or with means reported for each of multiple years, contributed more than one observation to the dataset. [Note: Many of the studies included in this dataset were experiments. Experiments involve manipulations that may be more common in the experiment than in the world, so average percent tip across all the observations in an experiment may be biased. Therefore, a second measure of percent tip based on observations from the control condition of experiments rather than all observations from those experiments was also recorded.] Year. When it was reported, the year of data collection was recorded. If data was collected across multiple years, the temporal midpoint was recorded. In some cases, the current author did not have the exact year of data collection, but did have evidence that it was substantially earlier than the publication date of the article. In those cases, the year of data collection was recorded as the most recent date consistent with known facts. Otherwise, it was assumed to be one year prior to date of publication for articles and the same year as defense for theses and dissertations State. The state from which each observation came was recorded to label points in a graph with the purpose of highlighting the many differences other than year of data collection between observations. If not explicitly reported, state of data collection was inferred from the author’s affiliation or other information. Data from multiple states were recorded as national.