4 ms·

This graph only goes back to 2000 but even on that timescale current military spending is lower than past peaks: https://data.worldbank.org/indicator/MS.MIL.XPN

This graph only goes back to 2000 but even on that timescale current military spending is lower than past peaks: https://data.worldbank.org/indicator/MS.MIL.XPND.GD.ZS https://data.worldbank.org/indicator/MS.MIL.XPND.GD.ZS

SiempreViernes 4y ago
Well, 2009 was when the bulk of the impact from the financial crisis hit world GDP so the peak is mostly reflecting that GDP contracted while military spending was mainly flat. You can compare the percentage of GDP figure to this one showing military spending as percentage of all government spending, where it looks like we're spending less and less on weapons: https://data.worldbank.org/indicator/MS.MIL.XPND.ZS https://data.worldbank.org/indicator/MS.MIL.XPND.ZS So while total dollar amount is maybe a coarse measure, the GDP or the fraction worldwide government spending are so contingent on the details of literally every other human activity as to unusable for quick observations.
- stingraycharles 4y ago
  This is an excellent point! As someone who’s totally not a data scientist, what approach would you take to counter this? Assuming military budget planning has a window of a few years, would a moving average of GDP be an appropriate solution? Or am I thinking too simplistic?
  - KolenCh 4y ago
    Not directly addressing this, but perhaps the point is not to create a model to address every known limitations. No models are perfect. Points are made based on a model, and a limitation/assumption is pointed out. You need to judge given the assumption/limitation if the message made still some-what make sense. I mean something X being highest right now is a pretty clear statement, it is either true or false, if X is well defined. But then your problem is you want to have Z instead, which hopefully is accounted for every other effect except for the factor you want to measure, such as inflation, GDP, etc. But the more you add, you either makes your Z definition sensitive to these other factors you’re not interested in (such as a peak comes from the denominator, not the numerator), or we really admit we can’t calculate Z for sure, and is now a statistical model (any kind of smoothing including the MA you mentioned is to admit we don’t know the true state but can only observe it with noise.) Now the statement of “Z is highest now” is no longer a true or false statement, but a probabilistic one. Certainly then P(Z is highest) = 60% or so wouldn’t make much of a headline and can be confusing. So my point was rather than trying to create a perfect model (which you can’t, and what you learn from it won’t be clear as well), sometimes just deal with the statement and methodology and judge if it is a clickbait or has some merit of truth there.

SiempreViernes 4y ago

Well, 2009 was when the bulk of the impact from the financial crisis hit world GDP so the peak is mostly reflecting that GDP contracted while military spending was mainly flat. You can compare the percentage of GDP figure to this one showing military spending as percentage of all government spending, where it looks like we're spending less and less on weapons: https://data.worldbank.org/indicator/MS.MIL.XPND.ZS https://data.worldbank.org/indicator/MS.MIL.XPND.ZS So while total dollar amount is maybe a coarse measure, the GDP or the fraction worldwide government spending are so contingent on the details of literally every other human activity as to unusable for quick observations.

stingraycharles 4y ago
This is an excellent point! As someone who’s totally not a data scientist, what approach would you take to counter this? Assuming military budget planning has a window of a few years, would a moving average of GDP be an appropriate solution? Or am I thinking too simplistic?
- KolenCh 4y ago
  Not directly addressing this, but perhaps the point is not to create a model to address every known limitations. No models are perfect. Points are made based on a model, and a limitation/assumption is pointed out. You need to judge given the assumption/limitation if the message made still some-what make sense. I mean something X being highest right now is a pretty clear statement, it is either true or false, if X is well defined. But then your problem is you want to have Z instead, which hopefully is accounted for every other effect except for the factor you want to measure, such as inflation, GDP, etc. But the more you add, you either makes your Z definition sensitive to these other factors you’re not interested in (such as a peak comes from the denominator, not the numerator), or we really admit we can’t calculate Z for sure, and is now a statistical model (any kind of smoothing including the MA you mentioned is to admit we don’t know the true state but can only observe it with noise.) Now the statement of “Z is highest now” is no longer a true or false statement, but a probabilistic one. Certainly then P(Z is highest) = 60% or so wouldn’t make much of a headline and can be confusing. So my point was rather than trying to create a perfect model (which you can’t, and what you learn from it won’t be clear as well), sometimes just deal with the statement and methodology and judge if it is a clickbait or has some merit of truth there.