topic Re: Constrained variables query in Discussions

Constrained variables query

NickT — Sat, 10 Jun 2023 23:30:15 GMT

Hi,

Can anyone advise a simple way to indicate that either a factor or an effect is constrained within limits ? Examples would include numbers expressed as %'s which must therefore lie between 0 & 100, such as Purity or the Yield of a reaction. This would be especially useful if it can be applied to a data table column, for example, rather than needing to be built into a custom design, thus enabling the simpler methods of fitting models such as Full Factorial or Response Surface

Thanks

Nick

Re: Constrained variables query

txnelson — Sat, 15 May 2021 23:30:56 GMT

Check out the column property, "Response Limits".

Help

and then search on Response Limits

Re: Constrained variables query

Mark_Bailey — Mon, 17 May 2021 18:41:49 GMT

Or maybe the Range Check column property?

Re: Constrained variables query

NickT — Wed, 19 May 2021 06:41:36 GMT

Thanks for the suggestion, but according to the Help entry for Range Check, this simply alerts to (or excludes) numbers that lie outside a predefined range for data entry This is fine for alerting to a typo or an excel formula error (for example) but does not reflect the position from a statistical analysis perspective. ie the model will still be constructed in terms of predictions below 0 or above 100 (% purity), even though such values these are meaningless numbers, nd cannot be attained in practice (other than by measurement uncertainty) - just as length is a continuous variable, except that its impossible to have a negative length.

Re: Constrained variables query

NickT — Wed, 19 May 2021 06:54:14 GMT

Thanks for the suggestion. I have taken a look at the options available from the drop-down menus in the Column Properties field (there are a lot of them !). Also I have referred to the Help. At least from the latter perspective the only clear instruction (to me) is that this is a way of setting the desireability of an outcome in terms of High and Low values. So, whilst this might reject an outcome of -1 or +101 as being unfit to meet the targets set, it would not actually influence the modelling process that could generate such numbers, I am no statistician, but I can remember from courses I took long ago that there are "special" ways of treating variables which are "bounded" - so whilst for example you might expect to find a Gaussian distribution of values either side of (for example) a solution being made up with a target of 50% concentration, if the target is changed to 100%, the distribution curve would look like one half of the normal bell curve, as its impossible (other than by measurement error) to get a value above 100%.

There may be something yet to be found in JMP on this matter, but frankly I don't find the Help very helpful - its good to have the Community as an alternative, so thanks again for pointing out this option.