Hacker News new | ask | show | jobs
by TeMPOraL 1338 days ago
> Eg. I've seen people convince a public company executives to do strategy X based on analysis Y. Then when things aren't going as promised I get asked to review the strategy, I might find some elaborate model where somewhere deep in the layers of math/logic/lookups, someone divided instead of multiplied or something like that.

Is it Excel-specific though? I've made these kinds of errors with Excel (though not with stakes this high). I've also made them a lot when doing math on paper. And I've also made them in C++, Common Lisp, Matlab, Python, R and JavaScript. Now, with those other tools, it's easier to spot an error in a formula on review - but in Excel, it's easier to spot the intermediary results being off, so it's a wash.

I think the thing to learn is to be more careful, to sanity-check intermediary calculations; these kinds of errors are about being momentarily confused, and will happen regardless of the tool you use.

2 comments

Yeah I agree with your conclusion and excel just happens to be the tool a majority of people are using so it’s a poster child of sorts. It also gives people “false” confidence in their ability at times when really they should probably ask for help. Using the reality of my team as an example again, a lot of people will just stay out of excel and ask us to build a model up front for a new strategy. My team isn’t flawless but much more experienced in the quirks and mental model of how to setup data for formulas, scenarios, etc. We also check each other’s work before something high stakes goes out of the team. Since we’re the experts it hurts us that much more when some miscalculation gets out and causes disruption.
> sanity-check intermediary calculations

I am very far from being an Excel guru, very far, but one thing I have found useful is having everything come out to an intermediate result, and calculate against those intermediate results. If you see weird numbers in the middle, then end result is likely wrong, figure out why your intermediate results are messed up. If you want to be fancy, you can even error bound your intermediates if you know your valid ranges.