• dogsoahC@lemm.ee
    link
    fedilink
    arrow-up
    11
    ·
    3 months ago

    There are a number of normalization algorithms. Easiest would be to just divide by the area’s population count. That gives you the relative number of bigfoot sightings or fursuits per capita, removing any skews introduced by varyin population size.

    Say you have two areas:

    Area 1: 100000 people, 1000 fursuits, 500 bigfoot sightings Area 2: 1000 people, 10 fursuits, 5 bigfoot sightings

    Without knowing the population size, it looks like more fursuits means more bigfoot sightings. But if we divide by the population size, we get 0.01 fursuits and 0.005 bigfoot sightings per person in both areas.

    Hope that helps. ^^

    • Fermion@feddit.nl
      link
      fedilink
      arrow-up
      5
      ·
      3 months ago

      Simple normalization does amplify signals in low density areas. If a person in a tiny town of 100 reports a bigfoot sighting and another person in an area with 10,000 population also reports a sighting, then with simple normalization the map would show the area with 100 people having 100 times as many big foot sightings per capita as the area with the population of 10k. Someone casually reading the map would erroneously conclude that the tiny town is a bigfoot hotspot and would in general conclude bigfoot clearly prefers rural areas where they can hide in seclusion. When the reality is that the intense signals are artifacts of the sampling/processing methods and both areas have the same number of fursuit wearers.

      • dogsoahC@lemm.ee
        link
        fedilink
        arrow-up
        3
        ·
        3 months ago

        That’s the point. To make the low-population area more intense. Because relative to the population density, there were 100 times as many sightings. Or what am I missing.

        • Fermion@feddit.nl
          link
          fedilink
          arrow-up
          3
          ·
          3 months ago

          I’m not saying normalization is a bad strategy, just that it, like any other processing technique comes with limitations and requires extra attention to avoid incorrect conclusions when interpreting the results.

          Because relative to the population density, there were 100 times as many sightings. Or what am I missing.

          If you were to attempt to trap and tag bigfoots in both areas, would you end up with 100 times as many angry people in a gorilla suit in the small town? No. You would end up with 1 in both areas. So while the tiny town does technically have 100x the density per capita, each region has only one observable suit wearer.

          Assuming the distribution of gorilla suit wearers is uniform, you would expect approximately 99 tiny towns with no big foot sightings for every 1 town with a sighting. So if you were to sample random small towns, because the map says big foots live near small towns, you would actually see fewer hairy beasts than your peer who decided to sample areas with higher population density.

          If we could have fractional observations, then all this would be a lot more straightforward, but the discrete nature of the subject matter makes the data imherently noisy. Interpreting data involving discrete events is a whole art and usually involves a lot of filtering.