On the up

14.03.2024 14:51 UTC+01:00
[ perma(?)link ]
< previous | next >

It's not what you say, it's the way that you say it. And that applies just as much to the mainstream media announcing Oscars 2024's viewership, which increased to 19.5M over last year's 18.8M. CBS News chose to celebrate this achievement with the following headline: 2024 Oscars ratings reveal biggest viewership in 4 years. Similarly, The Hollywood reporter went with TV Ratings: Oscars Score Post-Pandemic Highs, while those excitable little monkeys at TheWrap stuck their necks out and proclaimed Oscars 2024 Hit 4-Year Viewership High With 19.5 Million.

Not all outlets were as bullish, however. Variety led with 2024 Oscar Ratings: Academy Awards Audience Rises Slightly to 19.5 Million Viewers, and The Guardian with a sombre Oscars 2024: ratings up just 4% despite blockbuster Oppenheimer victory. Whichever way you look at it though, up is always up, even if it's only by 0.7M, or 3.7%.

Every silver lining has a cloud

chart showing Oscars' ratings from 2001 to 2024 — Oscars viewership: 2001–2024 (top), and 1974–2004 (above), showing the year-on-year declines during 1983–1987 and 2014–2017.

chart showing Oscars' ratings from 1974 to 2024 — Oscars viewership: 2001–2024 (top), and 1974–2004 (above), showing the year-on-year declines during 1983–1987 and 2014–2017.

On the downside, this year's viewing figure rather fucks with my model, such as it is, which predicted 13.2M for this year and now predicts 13.0M for 2025. It's worth noting that, although the viewership is the highest in four years, the year-on-year increase has declined steadily since 2022. This may represent an overcorrection from the pandemic crash that's now calming down. Time will tell.

On the other hand, the continual decline in viewership from 2014 (43.7M) to 2017 (32.9M) is not unprecedented. A similar pattern was seen between 1983 (53.2M) and 1987 (37.2M), the difference being that it reversed in 1988, from which point viewing figures remained consistently above 40M for the next decade-and-a-half.

So, perhaps the true decline in recent years is from 2017, not 2014. A regression model based on this tenet would give less weight to the higher viewing figures before the fall (Oscars2017, below). It's less robust, because there are obviously fewer data, and it still underpredicts for the last two years, but a little less so.

The data were modelled in STATA as viewers and year, from 2014 onwards but omitting 2021, using the regress and predict functions. The next year is appended to the year column beforehand.

Model: Oscars2014
regress viewers year
predict v

Model: Oscars2017
regress viewers year if year > 2016
predict v if year > 2016

The values for v through to current year are the trend, and that for the next year is the prediction. Not too sophisticated, huh? The correlation coefficient and its probability value can be taken from the regression table, but also from pwcorr:

Model: Oscars2014
pwcorr viewers year, sig

Model: Oscars2017
pwcorr viewers year if year > 2016, sig

		Model: Oscars2014				Model: Oscars2017
Year	Viewership	r	p	Predicted	Δ (%)	r	p	Predicted	Δ (%)
2022	16.6	-0.9507	0.0010	17.6	+6.0	-0.7997	0.2003	19.5	+17.5
2023	18.8	-0.9725	0.0001	14.0	-25.5	-0.9347	0.0198	14.4	-23.4
2024	19.5	-0.9670	0.0000	13.2	-32.3	-0.9241	0.0084	14.3	-26.7
2025	–	-0.9525	0.0000	13.0	–	-0.8944	0.0066	14.6	–

Comparison of the two models' characteristics, and viewership (millions). The correlation coefficient (r) and probability (p) are explained here. Note that the correlation for the Oscars2017 model was not statistically significant at α=0.05 for the first prediction.