Spreadsheets.
Mar. 31st, 2004 07:22 am![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
swestrupA few years after VisiCalc, the worlds first electronic spreadsheet, first appeared in 1979 I gave it a try. Although I was only 15 at the time, and knew little programming, it was clear to me that the product was severely lacking in features. I made a list of a dozen or so features that I thought it desperately needed, and over the decades since, most of them have now shown up. We still don't have reconfigurable topologies or arbitrary cell placement (ie, not just columns and rows), but they weren't the most essential features missing. Now, I've just read a new article that mentions a desperately needed feature that, in all my years of thinking about the problem, never occurred to me. That feature is error analysis and propagation, coupled with probabilities. I never thought of these as been missing, since they are things (I thought) that a scientist would need, and a number cruncher would have no use for. Heck, for all I know the long dead and lamented TK!Solver did all this. It seemed to be a spreadsheet designed for scientists (which may be why it failed).
Well, it seems that spreadsheets are no longer just used for tallying up what is, as I assumed, but are now extensively used for prognostication and trend analysis. The fact that they don't have ANY features that support this makes me wonder why folks are using such a broken tool for this purpose. Sadly, the fact is that the people doing the prognostication probably have no knowledge of error analysis or probability distributions and so are in no position to realize they don't know WTF they are doing. Even IF the spreadsheets had these features, I'm having a hard time imagining them being used, because the required ideas do not fit into the heads of the people making the decisions.
Still, I think that if one did some very careful thinking and market studies, one could write a "trend analysis spreadsheet program" that was specifically designed to lure corporate decision makers into the use of probability distributions and error bars instead of simply taking average or median values. The system would have to have LOTS of built-in smarts to make up for the ignorance of the user, but with some careful UI work, it could secretly be a teaching system as well.
IF (big if) the product took off it would make some real money, and do incalculable good by teaching folks how to make realistic predictions from limited data. Its an idea worth pondering, but I'm not going to put in anywhere near the top of my stack, since its far from clear that what I would like to do is possible, or that I could convince corporate types that they want reasonable projections.
For now, I'm filing this as a project to persue AFTER I've make my million, so I don't mind so much if it doesn't work.
Oh, and a quick web search shows that TK Solver still exists! Apparantly its vanishment from all store shelves 15 years ago was more in the way of a retrenchment than a complete failure.
Well, it seems that spreadsheets are no longer just used for tallying up what is, as I assumed, but are now extensively used for prognostication and trend analysis. The fact that they don't have ANY features that support this makes me wonder why folks are using such a broken tool for this purpose. Sadly, the fact is that the people doing the prognostication probably have no knowledge of error analysis or probability distributions and so are in no position to realize they don't know WTF they are doing. Even IF the spreadsheets had these features, I'm having a hard time imagining them being used, because the required ideas do not fit into the heads of the people making the decisions.
Still, I think that if one did some very careful thinking and market studies, one could write a "trend analysis spreadsheet program" that was specifically designed to lure corporate decision makers into the use of probability distributions and error bars instead of simply taking average or median values. The system would have to have LOTS of built-in smarts to make up for the ignorance of the user, but with some careful UI work, it could secretly be a teaching system as well.
IF (big if) the product took off it would make some real money, and do incalculable good by teaching folks how to make realistic predictions from limited data. Its an idea worth pondering, but I'm not going to put in anywhere near the top of my stack, since its far from clear that what I would like to do is possible, or that I could convince corporate types that they want reasonable projections.
For now, I'm filing this as a project to persue AFTER I've make my million, so I don't mind so much if it doesn't work.
Oh, and a quick web search shows that TK Solver still exists! Apparantly its vanishment from all store shelves 15 years ago was more in the way of a retrenchment than a complete failure.
