Issues with bivariate statistics variables

[prismv]

I tried to implement as RPL expressions standard deviation of linear regression results as sLR vector.
Furthermore In the HP50g user guide (hp50g-ug-en.pdf) page 18-52/-53 there are formulas of confidence intervals - cfLR is the vector name here.

Input was merely possible as a direct expression (or more precise multiplication of 2 expressions), because e.g. (ΣSize) is needed in input mode for proper evaluation, but the brackets are removed on saved expression on stack. There seems to be also a clash with ΣX² and (ΣX)².

The workaround was to use local variables for the dubious acting parameters.
The following shown implementation works for cfLR without any bad indications, but sLR gets a wrong output every second time Correlation is evaluated in alternation!

	dStats
		Directory {
			cfLR
				«
					'ΣSize' 'ΣX' 'ΣX²'
					'ΣY²' 'Correlation'
					→ N SX SXX SYY
					r
					«
						'√((N-1)÷(N-2)·SYY·(1-r²))'
						[ '√((N+SX²÷SXX)÷N²)'
							'1÷√ SXX' ]
						× →Num
					»
				»
			sLR
				«
					'ΣSize' 'ΣX' 'ΣY'
					'ΣXY' 'ΣX²' 'ΣY²'
					→ N SX SY SXY SXX
					SYY
					«
						'√((SYY-Slope·SXY-Intercept·SY)·SXX÷(N-2)÷(N·SXX-SX²))'
						[ 1 '√(N÷SXX)'
						] × →Num
					»
				»
			ΣParameters
				{ 1 2 49/5 1 LinearFit
				}
			ΣData
				[[ 1 11 ]
				  [ 2 19 ]
				  [ 3 33 ]
				  [ 4 39 ]]
		}

Because of several other reference implementations I can be sure that for given ΣData the result of sLR has to be [ 2.846 1.0392 ], so Slope is 49/5 with std.dev. of 1.0392.

Note: The result of cfLR with [ 6.838 1.8441 ] is not a fully expected result, because it cannot be calculated from the std.dev. values for intercept and slope by a single factor. So either there is still a typing error in my code, or an error in the HP50g user guide. But in effect this function showed me also the side effect of Correlation call on the other function sLR.

[c3d]

Thanks for the resport, @prismv. I will look into it.