Were there a posthumous prize for history’s most important poorly known scientist, the first recipient should surely be Sir Ronald Fisher (1890 — 1962).
It was Fisher who merged Mendel’s genetics with Darwin’s natural selection, creating our modern understanding of how evolution works.
In his spare time he invented modern multivariate statistics, including the analysis of variance, which, as it happens, is closely resembles natural selection.
Multivariatism is, I’m starting to think, a significant reason to embrace a principle espoused in The Cognitive Enterprise: Analytics are the new metrics.
Start with measurement. Measurement is the raw data. Ignore raw data’s detractors. It can have direct value, as, for example, your gas gauge telling you you’re low on fuel.
The popular “If you can’t measure you can’t manage” was never entirely true. You could, for example, drive forever without any of the instruments on your dashboard. You would, however, find yourself tanking up more often than you have to, just in case. You’d change your oil more often than necessary for much the same reason.
With nothing to estimate your velocity on beyond your perception of how fast the landscape is rushing by in the other direction you’d probably drive more slowly than your speedometer-enabled habits allow.
If you can measure, you can drive more effectively … less punchy but more accurate.
Measurements are numbers. Metrics are ratios. If you’re a car owner, your most important metrics are probably miles per gallon, miles per hour, cost per mile, ethanol as a percent of blood volume (I hope not), and other ratios that tell you how you and your car are performing.
If you’re obsessive about car care you’ll chart your automotive metrics over time to see if there are any trends. Changes to your mileage or operating costs over time might let you know of developing problems. So, for that matter, might your top speed (miles per hour), if you could gauge it safely.
Except that this is how metrics-obsessed business managers can get into trouble. Metrics report what. They don’t explain why, but not all managers care about subtleties like this. Something is wrong, which means we have to hold someone accountable.
For these fine managers, if their car’s mileage is deteriorating, their spouse, teenage offspring, or both are subjecting it to jackrabbit starts, driving way over the speed limit, or putting cheap gas in the tank.
Or, putting automotive analogies aside, if IT spending per employee is going up, IT must be buying technology for technology’s sake, business managers must be asking for new laptops for employees before they need to be retired, or the company should be negotiating harder with its IT vendors.
Or, sales are down (the metric: revenue per employee). Here’s an outstanding example of a principle I just made up: Metrics expand your opportunities to mismanage.
Because metrics report symptoms. They don’t diagnose.
To manage you don’t need to measure and you don’t need metrics. You need demonstrable causal relationships (hence “root causes”). You care about what the buttons and levers are that you can push and pull to change the metrics that matter for the better.
That’s buttons and levers in the plural. Very few business metrics change due to a single root cause. More often, several different factors interplay to cause the problem.
Most businesses are complex systems that operate in marketplaces that are also complex systems.
Which means business success or failure are multivariate affairs, which is why Sir Fisher earned a mention at the top of this column. Metrics tell you what’s changed. Understanding why and what you can do about it calls for high quality data and lots of it, applying multivariate analysis to untangle the multiple factors that cause it.
Not that this will deliver definitive results. Analytics give you correlations, which as we all know don’t prove causation.
Except they do, sort of, although “prove” is too strong a word.
If there’s a statistically significant correlation between A and B, the smart money says one of three conclusions is true: A might cause B, B might cause A, or there’s a C out there somewhere that causes both A and B.
Only it’s multivariate, so if G is headed in the wrong direction, and A, C, and F are all positively correlated with it … see A and B, above, only more so.
Likewise if G is headed in the right direction, except that no matter how strongly A, C, and F are correlated with G, DON’T TOUCH ANYTHING!
The root causes of success are just as hard to determine as the reasons for failure. Addressing them, on the other hand, is considerably more risky.
Not to mention less urgent.