It's a known fact that Robert McNamara was a huge proponent of using mathematical data analysis and computing to influence his decisions as Secretary of Defense. But is it known what calculations he did in particular, what his formulas and models were and to what extent they influenced his strategy?

by redditsuggsazz

I want to write an essay on the flaws of McNamara's methods (EDIT: In regards to the Vietnam war) for my history classes but I am finding that concise and accurate metrics in this regard are very difficult to come buy. The best I could find was a paper by historian Donald Fisher Harrison. But this is a professional paper and it is daunting for a novice like me. Are there any fundamental facts or metrics that would make this paper easier to digest?

flamefoxx99

I'm not going to do your homework for you, but I can maybe point you in a few directions. I'm a computer scientist by trade, so perhaps I can give you a few pointers.

First, foremost, and most importantly: data science is so much more than the underlying algorithms. The way you convert real world information like "the M16 tends to jam a lot" into a data point is objectively more important than the computation you execute. A statistician needs to define policies and mechanisms that dictate a) what data to observe b) how to observe it and c) how to interpret it; indeed: 80% of a Data Scientist's job is data collection and data cleaning.

If your paper does not give due diligence to the way that McNamara's statisticians interpreted real life data, it is fundamentally flawed.

For instance, PPBS, or Planning, Programming, and Budgeting System (First introduced in 1961 and largely designed by RAND Corporation) was a computational model used to optimize military that, in some respects, was the centerpiece of McNamara's computational revolution. This 1969 RAND white paper details the design elements of the PPBS. You'll notice how much work goes in to explaining the operation and policies in the model. Closer to the point, 1975 report on the systems analysis approaches used in the Vietnam War which focuses on the Hamlet Evaluation System, one metric used to inform and evaluate military strategy, has entire pages dedicated to the worksheets that NCOs had to fill out and send up in their reports. One worksheet reduces the disposition of a hamlet into an 18-point rubric. The construction of this rubric is just as important as the methodology used to evaluate it.

Second: computer modelling is often just about how you represent the data. A pretty important piece in Data Science history is Charles Minard's Flow Map of Napoleon's Russian Campaign. He includes no derived statistics, but this one map tells more than any individual statistic can. The brain is inherently a spatial computation engine orders of magnitudes more powerful than anything we can develop. Computer scientists at the time were well aware of this, and a lot of work went in to developing software that could simply ingest data and make it visually comprehensible to decisionmakers.

In scanning the same 1975 report above, I saw only one named computer program: Harvard SYMAP. This was an early GIS or Geogrpahical Inforomation System that essentially colored areas on a map according to some information associated with a certian data point. It basically made fancy heat maps. The underlying math for these involves Voronoi Diagrams, which you can probably find historical implementations for if that is important to you.

Third: the McNamara statistical revolution is inextricably tied to the RAND corporation. The RAND corporation will tell you this, they give this wonderful fluff piece detailing their contribution, but it's an excellent start.

Finally: I can't give you specifics on the underlying computational paradigms used, but this RAND report (which I can't find a good date on) details a history of RAND computations algorithmic approaches.

Key among them, that I would recommend as critical search terms:

  • Linear Programming: this is probably the most critical innovation. This is the breakthrough in computational optimization problems. You define the variables that you can change: "the number of tanks we generate this month, how many we allocate to this division, how many shells we give each tank", the constraints: "we can only support 10,000 tanks, and we can't give this division more than 2x the number of tanks sent to this division or else there will be political infighting", and an objective function which is a metric of your success "number of tanks sent to the 101st airborne", and it'll give you a solution that will optimize your objective function.

  • Monte Carlo simulation: ELI5: define a little system that will approximate the question you want answered. Run that system 1,000,000 times and you'll probably get something close to the right answer.

  • Game theory: I'm less familiar with this, but it was critical in dictating strategy

  • Computational Optimizations: stuff like dynamic programming and approximation techniques made it possible to compute problems that would otherwise take millions of years in the span of minutes.

I enjoy the history of computer science, and this was fun. If you need any help understanding jargon or any computational technique, feel free to message me.