Thinking Clearly with Data by Ethan Bueno de Mesquita & Anthony Fowler

Author:Ethan Bueno de Mesquita & Anthony Fowler [Bueno de Mesquita, Ethan & Fowler, Anthony]
Language: eng
Format: epub
ISBN: 9780691214368
Publisher: Princeton University Press
Published: 2021-11-15T18:30:00+00:00

What’s going on here? Is this correlation attributable to a true causal effect of science spending on suicides, to bias, or to noise? It’s theoretically possible, but very unlikely, that science spending has a large, positive effect on suicide by hanging (or vice versa). Noise certainly seems like a plausible explanation. If you look at enough variables, you’re bound to find two of them that happen to correspond by chance, and we know that this is exactly what Vigen did. He checked for correlations over time for many variables and selectively reported the correlations that were significant.

But maybe it’s also bias. What’s an example of a confounder here? Could there be a variable that affects both hanging suicides and also science spending? One potential confounder is population. Over this period (1999–2009), the U.S. population grew steadily from about 279 million to 307 million. And population growth could plausibly increase both suicides and science spending.

To explore whether bias or noise is the more important explanation for the observed correlation, it might help to think about whether you expect this correlation to also hold for years before 1999 and after 2009. If you suspect that this correlation would likely hold more generally outside this sample of data, then it can’t just be noise. Alternatively, if you think that this correlation is just a fluke, unlikely to hold outside the short period for which Vigen collected data, then it’s just noise, due to neither a causal relationship nor bias.

Let’s take a look at another couple examples. Figure 9.11 shows the correlation over time between sociology doctorates awarded in the United States and worldwide non-commercial space launches. Again, there’s a strong correlation. Furthermore, it’s not so easy to simply attribute the result to population growth (or something else changing over time) because the correlation is not driven by the two variables generally increasing over time. On average, space launches and sociology doctorates aren’t increasing or decreasing, but the years with more space launches also tend to be years with more sociology doctorates.

We’re pretty comfortable chalking this one up to noise. There’s idiosyncratic variation from year to year in space launches and sociology doctorates, and they happened to line up during this period. But we suspect that if we looked at the next thirteen years of data, the correlation would be close to zero.

Figure 9.12. The metaphysically challenging correlation between Nicolas Cage movies and swimming pool drownings.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.