Monday, January 19, 2015

Data Worship and Duty

If you spend more than a few minutes working as an analyst--operations, program, logistics, personnel, or otherwise--it is almost inevitable that some wise military soul will offer trenchant historical lessons about undue trust in analytics for decision making derived from the performance of Robert McNamara as Secretary of Defense. Too often, these criticisms are intended to deflect and deflate criticisms and conclusions of analysis without addressing the analysis itself (an ad hominem approach without so much of the hominem). But that doesn't mean there aren't common mistakes made in the conduct of analysis and worthwhile lessons to be learned from McNamara.


This short article from the MIT Technology Review is a bit old, but it also makes a number of useful points. The "body count" metric, for example, is a canonical case of making important what we can measure rather than measuring what's important (if what is important is usefully measurable at all). Is the number of enemy dead (even if we can count it accurately) an effective measure of progress in a war that is other than total? So, why collect and report it? And what second-order effects are induced by a metric like this one? What behavior do we incentivize by the metrics we choose, whether its mendacious reporting of battlefield performance in Vietnam or the tossing of unused car parts in the river? 

There's something more fundamental going on in the worship of data, though. We gather more and more detailed information on the performance of ours and our adversaries' systems and think that by adding decimals we add to our "understanding." Do we, though? In his Foundations of Science, Henri Poincaré writes:
If we could know exactly the laws of nature and the situation of the universe at the initial instant, we should be able to predict exactly the situation of this same universe at a subsequent interest. But even when the natural laws should have no further secret for us, we could know the initial situation only approximately. If that permits us to foresee the subsequent situation with the same degree of approximation, this is all we require, we say the phenomenon has been predicted, that it is ruled by laws. But this is not always the case; it may happen that slight differences in the initial conditions produce very great differences in the final phenomenon; a slight error in the former would make an enormous error in the latter. Prediction becomes impossible and we have the fortuitous phenomenon. 
Poincare is describing here what would later be dubbed the butterfly effect for nonlinear systems (with the comparison to predicting the weather made explicit in a later chapter). In systems such as these, chasing data is to pursue a unicorn and the end of the rainbow. Rather, it is structure we should chase. Modeling isn't about populating our tools with newer and better data (though this may be important, if secondary). Rather, modeling is about understanding the underlying relationships between the data.

We often hear or read that some General or other should have fought harder against the dictates of the McNamara Pentagon, but one wonders if perhaps such a fight is also the duty of a military analyst.

10 comments:

  1. James Hanford
    (Reposted from Facebook, with permission)

    "We gather more and more detailed information on the performance of ours and our adversaries' systems and think that by adding decimals we add to our 'understanding.' "

    Well put. We tend to conflate data with truth, mistaking precise expression of an aspect of reality for reality itself. More data = more truth by default for many folks, as they miss the point.

    Math and data aren't the same thing - though I wonder how many folks confuse the two. Anyway, it reminds me of the apocryphal story of Euler and Diderot (http://www.fen.bilkent.edu.tr/~franz/M300/bell2.pdf) trying to use math to make a larger claim, though in this case it was to embarrass one side. I more focus on the embarrassment held by Dierot when there really shouldn't be much embarrassment (nor joy of the spectators who misunderstand the use of math and probably would be overenthused with precise data). I guess if you call this "Eulering" someone then we tend to Euler ourselves quite a bit. We do like our data...(or clever rhetoric).

    ReplyDelete
    Replies
    1. I've always rather liked the story of Euler and Diderot for a variety of reasons, not least because it provides a lovely point of departure for any number of interesting excursions. But this is not the first metaphysical argument (or story of such an argument) over mathematics. There is a great story that Hippasus was drowned by the gods (or set adrift by the Pythagoreans) for proving the existence of irrational numbers, for example. We take our math seriously here.

      Delete
  2. Dave "Sugar" Lyle
    (Reposted from Facebook, with permission)

    Nice, Merf. Agree that it is an ethical issue to be up front about what our analytics can tell us, and what they can't - scientists experience the same social pressures and biases that the rest of us experience, especially if their identity, grants, and reputation rest upon certain ideas, tools, and methods. But that applies equally to those who use the analysis, who need to meet the analysts halfway to get the most we can out of the tools we have, and to avoid making claims greater than the analysts are making (the sin of many science writers and CEOs)

    ReplyDelete
    Replies
    1. I don't disagree with you in the least. Decision makers and analysts (and everyone else, for that matter) have obligations in this dialogue. McNamara's tragedy was to be at the center of a perfect storm of decision maker and analyst that each reinforced the beliefs of the other and ran about unchecked. The particular slant of the short post I put up is driven by the intended audience for that forum ... it was originally intended for "discussion in the analysis realm," so I'm talking primarily to analysts.

      Delete
    2. Dave "Sugar" Lyle
      (Reposted from Facebook, with permission)

      http://news.stanford.edu/news/2015/january/conflict-felter-database-011615.html

      Delete
  3. Doug Kupersmith
    (Reposted from Facebook, with permission)

    Metrics are easy. Meaningful metrics less so. Credible, decision-quality metrics are pure gold. There are those who have a nose for the gold and there are those who settle for plain metrics, because they can. It behooves anyone whose work touches the military OR business to have a good feel for the full continuum of metric usefulness so they may recognize the good and the bad for what they are.

    P.S. MacNamara's photo still makes me shiver inside. In 1994 I still had to sit through Glen Kent diatribes about what a criminal that man was.

    ReplyDelete
    Replies
    1. Dave "Sugar" Lyle
      (Reposed from Facebook, with permission)

      Doug, I don't know if it would be therapeutic, or just make it worse, but have you ever seen this retrospective?

      https://www.youtube.com/watch?v=nU1bzm-BW0o

      Delete
    2. Doug Kupersmith
      (Reposted from Facebook, with permission)

      McNamara's autobiography (1996) carried the same tone as that movie trailer: War is really really hard to understand so mistakes get made. I guess he forgot the part where those more ignorant of war shout down those with the knowledge, experience, and insight to deal with it more successfully. I wonder when Cheney and Wolfowitz plan to publish their thinly veiled mea culpas? (OK, that's easy for Cheney: never)

      Delete
    3. The simple metric that reliably points north across a variety of conditions is gold indeed. So, here's a question ...

      What are some examples of metrics in each category? We already have body count, and I think we agree that is a metric of the first type (easy). So how about meaningful metrics? Credible, decision-quality metrics?

      Delete
    4. Doug Kupersmith
      (Reposted from Facebook, with permission)

      I’m slow responding to your request for good metrics examples. I have 2 - a successful one and an untested one.

      The successful one came from an early 1990s study comparing the F-15 to the F-22 (congressionally mandated silliness, but a story for another time). In short, it was hard to show a significant difference between a current fighter and a future fighter using the old tried-n-true loss exchange ratio (LES) especially since the current fighter had a real-world LES with a zero in the denominator (100+-to-nuthin’ and counting). So we focused our metrics on differences between the 2 fighters, which fell into four broad (and unclassified) categories: Stealth, maneuverability, speed (actually acceleration rate), and avionics. What resulted was a metric, in graphic form, comparing a common bad guy airplane against the F-22 or F-15, by depicting relative engagement potential of the opposing aircraft. The graphic looked innocent enough, but the comparative lethality, vulnerability, threat avoidance, and even LES were all mathematically displayed (though clever use of geometric areas/shapes) in an easy to understand cartoon. Dr. BS Sheldon and Milt Johnson were the main drivers in this.

      The untested one I’m still waiting to see, although I’m less involved with the community of late so this may be dated. It centered on ISR and the seeming inability of that community to generate a compelling yet easily understandable metric (circa 1990s – don’t hate one me yet Mooch!). Mired in the “if I make more observations I’ll know more” mentality of the time, the community seemed unable to answer a basic “so what?” question (i.e. what is the effect of this on operations?). My suggestion seemed simple on the surface (for the intended audience of decisionmakers) but would be difficult to obtain – a numeric assessment of total situational awareness. I called it: What I know about what I need to know. First, what planners/battle managers “need to know” is quantified in a core metric. Result: a lot, but not everything. In theory, this is what they do for a living, they just don’t tend to routinely quantify it. Next, what is “known” is quantified in a separate, basic metric. From these two metrics come three excellent insights: 1. What is my % of total SA (ISR effectiveness)?; 2. How do my need to know and known metrics change with respect to each other over time (ISR demand/supply)?; and 3. How much do I know that I don’t need to know (ISR/C3 inefficiency). Everything else is rubbish.

      Lastly, Dan Hackman is a visualization guru who repeatedly produces unique and pertinent metrics for all those crazy studies he gets pulled into. I know BS, Milt and I are all left-handed. I believe Dan is too. Now there’s a population sample for you!

      Delete