Skip to content

Smiley faces were never seen

Jay Livingston shares a graph from this paper by Shiry Ginosar, Kate Rakelly, Sarah Sachs, Brian Yin, and Alexei Eforos:


The graphs summarizes an analysis from a database of high school yearbook photos.

Livingston writes:

Ginosar et al. have only one explanation for the upward trend – technology. In the early 20th century, they say, photo portraiture was still under the influence of 19th century technology. Those old cameras required an exposure of several seconds, sometimes as long as half a minute. When you have to be motionless for that long, a neutral expression is easiest to maintain. . . .

The trouble with this explanation is that the Kodak camera was introduced in 1888. By 1900, everyone was taking snapshots rather than posing solemnly for photographs taken by a man hiding under a black cloth with a large wooden box on resting a tripod. The snapshot was to 1903 what the selfie was to 2013. But perhaps old poses hang on even though they are no longer technologically necessary, and fashions in yearbook poses diffuse gradually.

He continues:

But why the decline in smiles from 1950 to 1965? These were, by some accounts, the most contented years of the century, free of conflict and turmoil, even boring. . . . I [Livingston] have no idea. You lovers of zeitgeist explanations, feel free to speculate.

I wish they’d shown the data for every year. The patterns are smooth enough that it seems wasteful of information to bin into 5-year categories.


  1. Z says:

    Speculation 1: The more frequently you’re photographed, the more practice you get at smiling on command.

    Speculation 2: Maybe dental care has improved and people are less embarrassed by clunky braces or crooked teeth?

  2. Steve Laniel says:

    It’s so uncommon for anyone to paraphrase “We Work The Black Seam” — which is an excellent song — that I feel it necessary to stop here and tip my cap to you.

    That is all.

  3. Tom Passin says:

    It’s pretty clear from the graph that there has been essentially no change since at 1950 if not before. At least, with error bars like that, there is no support for changes. If the error bars are 2-sigma, you could even make a case that there has been no serious change during the whole time, though I suppose realistically there really were probably straighter lips in the early part of the century. [Who wants to work up a p-value on that??]

    To the extent that there was any change, even if you could take a picture with a short exposure in the early 20th century (as Livingston remarks), people back then still thought that you should be serious when photographed. Mark Twain is often quoted on this:

    “A photograph is a most important document, and there is nothing more damning to go down to posterity than a silly, foolish smile caught and fixed forever.”

    (See, e.g.,

    Smiling faces in photos weren’t all that common until more recent years.

    • Ben Bolker says:

      I don’t know how much reading of TFA should be expected before commenting, but the figure caption clearly states these are +/- 1 standard deviation bars. They used about 38,000 photos over the course of a century (don’t know exactly how they were distributed across time), so approx. 380 photos/year, so the standard errors would be ~20x smaller than the standard deviations …

      • Elin says:

        Yeah at first I thought “what?” and then I realized it wasn’t the standard error. The thing is … if they are saying there is essentially a linear trend they should model that rather than act like each year is independent of each other year.

    • bxg says:

      If nothing else, the correlation between male and female is fairly convincing support of the apparent changes IMO, even post 1950. If we were looking at “basically constant, plus noise” why would two disjoint samples per period be so consistently correlated?

  4. Martha (Smith) says:

    The data set used could conceivably affect the graph. On p. 2 of the paper, the authors say the pictures come from “128 schools in 27 states.” Figure 2 of the paper shows that the distribution by region varies from year to year. For example, the number of yearbooks from midwestern states appears to have a trough from about 1948 to 1958. So comparisons between years are questionable.

    • Elin says:

      I don’t think there’s any question that a more complex analysis would really be necessary if they were publishing an article that was more about the results and less about the technology they used to collect the data. I mean if you were really going to model it you would need to deal with the nesting of images in schools as well as whether some schools were present multiple times.

Leave a Reply