7
$\begingroup$

I have a conspiracy-prone friend who believes that people are controlling the weather to ensure good weather for a large, annual festival (last weekend in June in San Francisco). As a scientist, (not earth science), I was interested in testing this hypothesis. So..I built 2 linear models--one representing the days leading up to the festival, one representing the days following the festival-- where the I regressed max daily temperature onto the number of days away from the festival. Lo and behold, indeed the temperature did appear to rise, peak on the weekend of the festival, and then drop afterwards! Next I figured, it must just be a natural trend, irrespective of local festivities (and weather-controlling overlords). So next, I included data from the years before the festival began (1972), made two groups (pre-festival and post-festival) and included a days_away_from_festival*group interaction term. To my great surprise, the interaction was significant! Testing only the group dated before the festival began, did not yield significant results.

And so! My question is... can any of you experts explain this by means other than a statistical coincidence/wealthy weather gods? Why might the temperature in San Francisco tend to rise in the week leading up to the last weekend in June, and then fall afterwards--but for this effect not to be present in the years before 1970'sish?

EDIT: Here are my two models.

The first model represents the days leading up to the festival and days are indicated by the countdown variable, (e.g. Friday = -1, Saturday = 0) There are 5 days prior included in the model. The group variable refers to the fact that I split the years into two groups, old:(1942-1971) and new:(1972-2017). TMAX is the maximum temperature of the day.

pre festival model

The second model is the same except it represents the days following the festival. post festival model

As you can see, after 1972 the temperature seems to rise, peak during the festival in the last weekend in June, and then fall immediately after. However, this effect was not present prior to 1972.

EDIT2: The data came from https://www.ncei.noaa.gov and the station is San Francisco Downtown

$\endgroup$
  • 2
    $\begingroup$ Two ideas. First, the station seems to have moved around a bit, including in 1973 (ncdc.noaa.gov/cdo-web/datasets/GHCND/stations/GHCND:USW00023272/…). Second, global warming could be a confounding signal over those decades, as could increasing urban heat island effect. If possible, see if you get similar results a) from other urban sites unrelated to the festival or b) from relatively nearby rural sites. $\endgroup$ – Jareth Holt Jun 29 '18 at 7:13
  • 3
    $\begingroup$ You do know that this is futile, right? It's a fine exercise for yourself, but you're not going to "prove a conspiracy believer wrong" with data. $\endgroup$ – Jan Doggen Jun 29 '18 at 13:38
  • 1
    $\begingroup$ Perfect example of Trump's Razor: Ascertain the stupidest possible scenario that can be reconciled with the available facts. Of course it's a conspiracy! $\endgroup$ – Tim Nevins Jun 29 '18 at 15:23
  • 1
    $\begingroup$ Did you consider that the data may have been manipulated for some inscrutable purpose? 🌎 $\endgroup$ – Keith McClary Apr 23 at 1:02
  • 1
    $\begingroup$ Although I'm not confident I fully understand your data analysis, the results sound exactly like what one would expect when testing a post hoc hypothesis developed from the very data one is using for the testing: the significance is grossly inflated and there is an interaction with a temporal indicator. If I'm correct, then you have little chance of changing your friend's opinion. You could take similar datasets for other cities, dredge them to develop comparable hypotheses, and perform post hoc tests to show you can always find a "significant" result. All cities control their weather! $\endgroup$ – whuber Apr 23 at 2:33
2
$\begingroup$

Let's have a closer look at the p-values in your analysis. By far the most significant effect in your model is the one captured by "groupold", which indicates a significant difference between pre- and post-1972 temperatures. This points to the climate change effect, as noted in John's answer.

The next most significant factor we have is "countdown", which indicates, not totally surprisingly, that temperatures tend to increase in the end of June and cool off into July.

Finally, we have the interaction term, which is orders of magnitude less significant than the other factors. The most reasonable explanation for this is that climate change effects are most notable in the extremes of weather. As climate change raises the overall average temperature, it also pushes the extremes higher and lower. Climate change has a significant effect on temperatures, but that effect is even more pronounced in the height of summer.

So no, the festival organizers haven't been using their WeatherSetter 9000 to make the last week of June warmer than average. Rather, climate change has made weeks since 1972 warmer than average, seasonal cycles have always made the last week of June warmer than average, and the climate change effect is especially strong at the extremes of weather.

$\endgroup$
  • $\begingroup$ Brilliant! This is absolutely the most plausible response, in my view, that I have received on this question. Thank you for satisfying my curiosity after 10 months of wondering 8-) $\endgroup$ – Ashish Apr 26 at 19:24
  • 1
    $\begingroup$ @Ashish It would be interesting to test this hypothesis for other locations/times of year, to see if the interaction between climate change effects and seasonal temperature cycles remains consistent. If my supposition is correct, we should see this effect occur in other locations, but be less pronounced during months of less extreme weather. $\endgroup$ – Nuclear Wang Apr 26 at 19:38
5
+50
$\begingroup$

You are asking the wrong question. The real question is why some podunk single city festival and not other larger more costly events, like space launches, national holidays, military campaigns, hurricanes, or a million other festivals. The conspiracy theorists likely picked that festival through data dredging, which means you are not going to disprove it by looking at the same data. If you compare enough festivals you will find some in which the weather does weird things, it is the law of large numbers in action.

A festival that started in the 1970's is perfect for data dredging, there is a change in the underlying forces of weather in the 1970's. The undeniable warming trend in global climate starts around the 1970's so if you divide your data sets before and after any point around that time the data will always show a significant difference, unless you correct for that shift. Without correcting for it all you are detecting is climate change, a known factor,n anthropomorphic factor at that. So really humans are altering the weather, just not in the way your friend thinks.

enter image description here

enter image description here

$\endgroup$
  • 1
    $\begingroup$ The question was asked specifically about the last weekend of June, temperature rising and falling. This is unrelated to the long-term trend. $\endgroup$ – Gimelist Apr 25 at 12:54
  • 1
    $\begingroup$ but you are noting how this changes over many years, you note the behavior of the weather changes in 1972 that is the significant factor. the fact the average temprature is rising will impact this. temprature rising and falling from day to day is completely normal. $\endgroup$ – John Apr 25 at 14:52
  • 1
    $\begingroup$ The long term trend is controlling the only significant factor, that the data changes behavior in the 1970's nothing else in noteworthy. the rise and fall is just statistical noise, it's not significant. it is pure data dredging. like drawing the bullseye around a cluster of shots already fired and declaring superb accuracy. $\endgroup$ – John Apr 26 at 3:52
  • $\begingroup$ There was no data dredging. I tested one hypothesis. Every decision I made was strictly driven by theory (misguided theory nonetheless, but still not data dredging). You make it sound like I was testing all different years to split arbitrarily. And secondly, the day-to-day rise and fall is not just statistical noise--that's precisely the point of the post and the slope was non-zero in the "conspiracy-ridden years" and 0 in the "innocuous years". If it were noise, it would cancel out to be a slope of 0 in all of the years. $\endgroup$ – Ashish Apr 26 at 14:15
  • $\begingroup$ And in regard to your first point, obviously this is a spurious effect. I am just trying to determine why I found it. This is not a common theory that conspiracy theorists found by data dredging. Weather manipulation in general is a known theory, but in regard to this festival this was just something that my friend with absolutely no data analysis skills came up with based on his anecdotal experience and memory. $\endgroup$ – Ashish Apr 26 at 14:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.