In GraphBuilder, for fit of line, is it possible to set the y-offset fixed to 0?
Hi @hogi ,
No, I don't think that it is. I think that you are asking to constrain the intercept to zero. For more complex modelling options like this, you need to use Fit Y by X from the Analyze menu. From the red triangle menu for Bivariate Fit you can select Fit Special... and then you can select Constrain Intercept to 0:
I hope this helps,
Phil
You can fit a model as @Phil_Kay describes, save the prediction formula, and add that column to the Y axis as a function graphics element.
Hm, I was searching for a one-click solution to get a "1:1" correlation line *) into the Graph Builder Plot, one which goes through 0/0 and runs close to the data points **) - I guess I will write a script which just calculates the center of gravity and draws the line.
*)edit:along the idea:- 0 % -> 0%- 100% -> 100%For all the cases where it's 100% clear that X and Y correlate without offset,
like a conversion of scales: any jitter around the line can just be a measurement artifact (either on the X or Y axis - or on both).
**) By the way - actually I thought that a fit with constrained Y intercept goes somewhere close to the center of gravity:no freedom for the Intercept, 1 degree of freedom to get close to the data points.
I was a bit surprised to see (red line: y not constrained, green line y=0)
So, calculating the center of gravity and drawing the line seems to be the better approach for what I want to do ...
When you say ""1:1" correlation line", do you want a 45-degree line i.e. y=x : if so, you can do that in Graph Builder
right-click> customise> add a script using the Y Function template and set the function of x as x:
Y Function(x, x);
(ok, not a "one-click" solution, but very effective if you want a reference line for a calibration curve, or a reference line for actual versus predicted)
Thanks I will use the Y function to add the line.But it will not be y = ONE times x.
Let's say X is measured in cm and Y is measured in inch.Same holds for measurements of the same property at different tools - or at different times during production process.(if I know that 0 is measured as 0)
Argh, I just thought again - and remembered a wonderful micro-"lecture" that @jules gave in the DataViz breakout room at Discovery Summit Europe 2023 - how easy you can solve puzzles with regression - by showing the squared residuals in the plot via
Open( "$SAMPLE_DATA/../Scripts/InteractiveDemos/DemonstrateRegression.jsl" );
After dragging the points and lines one can sees that my puzzles with the constrained fit were actually the ordinary thing with fit outliers - and the squared weighting of the residuals - and a fit line which definitely doesn't hit the center of gravity:
