Good Metrics Produce Numbers

The best metrics have a numerical result and an objective way to obtain the numerical result. However, this is only a prerequisite to a good metric, not a definition. For example, LinesOfCode is an objective metric, but not necessarily a good one, at least not by itself.

I am assuming that this is you, TopMind; if not, maybe there is some point to discuss here. At best, it says very little, so if you are not Topmind and wish to elaborate, please do... -- Bottom

[EditHint: merge with GoodMetricsUseNumbers. This topic name may be a better fit.]

Eesh. I really don't want to belabor the point, but doesn't the term "metric" imply the use of numbers? Isn't a metric a measurement? How else do you get a numeric measurement of something other than with, well, numbers?

"Kind of warm" could also be called a metric, but a messy one. Anyhow, what topic name would you suggest as a replacement?

Hmm. "Kind of warm" isn't a metric; it's an evaluation. Metrics are used to measure and quantify numeric values. If you double "kind of warm" do you get "a little too hot," "way too hot," or just "pretty warm?"

On the other hand, if you are using relative values for enumeration then perhaps "kind of warm" can be defined as halfway between "too cold" and "warm enough." If you define it that way then you have something to talk about. You can apply proper metrics to "kind of warm" and you know what you're measuring and what it means.

As for the page title, hmm. Is there some distinction that needs to be made between good and bad metrics? Don't they both produce numbers? Lines of code is a solid metric, even though it means diddly. Loadable object size is a solid metric, but it tells you almost nothing as well. Execution time is a metric that may have some value, as does latency, interrupt response time, yada yada yada. All these are metrics, and they all produce numbers.

Keep in mind being a prerequisite for "good" is not the same as being sufficient for "good".

Sorry, but once again we're getting away from the use of numbers and measurements. What is the value of "good?" How is it measured? In what units? See the discussion of "kind of warm," above.

Metrics involve measuring. If we can measure some factor and express it in numbers then we know something about it. Otherwise, we're just blowing smoke.

A survey of people ranking a software program design on a scale of 0 to 9 can be represented as numbers. But, that by itself doesn't guarantee a good metric is being used. If they are ranking it by how close it fits the latest IT fad, that doesn't say anything about the value of the fad itself, for example.

Correct. See the reference to lines of code, above. For any metric to have value one must establish the units of measurement. In your example the unit is a relative ranking. This measurement will have value to somebody, just as LOC will have value to some dork somewhere.

Create a measurement for some factor of known application and you've got something. Measure some factor that is highly subjective and relative only between comparable units and you have a measurement of very limited value. Apply some smarts.

"What is a good metric" is a wider topic than this. Producing a number-able result is merely a prerequisite. I vaguely remember some existing discussions on the nature of good metrics. I think I ended up concluding that in the end it's all relative (surprise surprise), creating the usual flame-wars. But even if it's relative, numbers provide a tool for evaluators to use to construct their own relative models for their own evaluations. They assist in the consistency within the relative model.

Measure what makes sense. Here we have a lead-acid battery. It weighs this much, has that much storage capacity, and delivers current at so-and-so a rate. Over there we have a lithium polymer battery that weighs what it weighs, has whatever storage capacity, and delivers current at yada yada rate.

These two batteries can be compared in many, many different ways -- the lead-acid isn't rechargeable, the LiPo is much lighter, the current dump capacity is different, the explosive potential with case penetration or over-voltage is different. All of these factors can be measured and have some meaning and value one way or another to all interested parties. However, the one measurement we can all agree on is that the lead-acid battery is 6 VDC and the LiPo is 7.4 VDC. There is no room for argument about that at all.

By establishing what it is we're interested in measuring before we get started then we have something to discuss. Once we can agree on the factor under scrutiny then we can settle on measurement criteria, units, etc. Until that is established then we're always going to have these useless, circular discussions.

All numbers are useful to somebody at some time. You gotta pick the numbers that are going to be most useful to your progress before you ever get started, or everybody is wasting his time. In the mean time, what's your battery voltage?


EditText of this page (last edited January 2, 2013) or FindPage with title or text search