Andrew,

Do you want to do this automagically or one trend line at a time interactively?  There are some ways to do this kind of thing, but the route depends on how the visualizations will be consumed or used.  Can you provide a bit more information?

MANY BLESSINGS!

Peace and All Good!

Michael W Cristiani

On Twitter

I wanted to bump this as well because I have a similar scenario.  I am trying to create an optimal pricing model so if I use the base trend line it's no good because my customers with poor margins bring down the trend.  I would like to be able to adjust and calculate a line and plug in that formula and at the minimum have Tableau draw it for me (at maximum be able to use it in calculated fields).  Right now I am having to export it out as an image and using a photo editor to draw the line. 

Brandon,

With custom table calculations, I believe what you are looking can be done, but you would need to provide sample data, and the formulas and/or business logic you want applied, and what you expect for a result. With additional details and specifics on what you have and what you are looking for, assistance can be provided.

Hi there,

I want to bump this thread - did any of you find a solution?

What I want to do is to "fetch" the last value of a trendline and use it in a calculated field.

As far as I can tell I would somehow calculate the field myself, if I know how the trendline is calculated, right?

So if I use "Describe the Trend Model" I get the "Model Formula" which I do not understand. It multiplies some values and then adds an intercept.

Is there a way to logically derive the necessary structure of a calculated field from a trend line? Where would I start?

Thanks,

Peter

Any news on this one? :)

So I reckon no?

Would it help to contact customer support?

Peter,

does: http://www.alansmitheepresents.org/2011/06/fitted-curve-modeling-in-tableau.html help your situation?

or are you looking for something like: The specified item was not found.

or do any of the comments at http://www.tableausoftware.com/support/forum/topic/forecasting-how address your situation?

If these do not help your situation, then if you can provide a sample packaged workbook that represents your situation and detail what you expect for a result, an example solution can be made.

Hi,

I saw one feature which helps for advanced analysis in the below link.

http://community.tableau.com/message/181143#181143

I think this feature can help us for doing prediction also.Am I correct?

Philip

Greetings,

There seems to be a lot of interest in being able to calculate the slope for a linear regression in Tableau above and beyond the built in trend line. This makes sense as sometimes you might want to sort or isolate things with certain slopes. I would give this calculation a shot:

(size()*window_sum(sum([x])*sum([y]))-window_sum(sum([x]))*window_sum(sum([y])))

/

(size()*window_sum(power(sum([x]),2))-power(window_sum(sum([x])),2))

Here your [x] field is your x axis and your [y] field is your y axis.

For time series it might be a little more difficult. If you have data by day, I might try making your [x] field something like: datediff('day',today(),[Date]) assuming you don't have data going out into the future. If you do, you can just anchor the datediff calc to a random date in the past before your data starts (datediff('day',date(1/1/1998),[Date])).

EDIT: or you can just be smart and create a field float([Date]) and use that instead

This should pull out your slope value. You can back into your intercept and rsquared with similar logic.

Remember that this is a table calculation so if you are looking to slice this up by a dimension will most likely need to edit your table calc to restart every [dimension] (and change the compute using).

Anyone finding this by searching and wants to see how to calculate slope of a trendline using LOD expressions:

I'm mimicking this formula to obtain slope: y = αx + β

where α = (n∑(xy) -∑x ∑y) / (n∑x^2 -(∑x)^2 )

({fixed:count([X])}*{fixed:sum([Y]*[X])} - {fixed:sum([X])}*{fixed:sum([Y])}) /

({fixed:count([X])}*{fixed:sum(power([X],2))}-power({fixed:sum([X])},2))

Where [X] is your x-axis, [Y] is your y-axis

If your x-axis is a date, create a calculated field of float([X]) called floatdate.  Substitute [X] in the formula with your new calculated field, like so:

({fixed:count([X])}*{fixed:sum([Y]*[floatdate])} - {fixed:sum([floatdate])}*{fixed:sum([Y])}) /

({fixed:count([X])}*{fixed:sum(power([floatdate],2))}-power({fixed:sum([floatdate])},2))

Obviously, if you need to edit the level of detail, add the necessary fields between 'fixed' and the colon.

Joey,  This method worked splendidly.  Thank you for your contribution to the community.

The only thing is that I think it does not handle situations where my LOD has NULLs, but that's on my end.

Joey,

I have a very similar case, but can't seem to get the formula to provide the right answer.

I'm attaching a sample .twbx

For the data that's in there, Tableau is calculating a slope of .224237 when I bring in a linear trend line. I want to create a calculated field that will return that same result, but I have been unable to get the same result using your guidance above. I think it might be related to fact that my Y value is already an AGG function. If you could take a look, I would really appreciate it. Let me know if you have any questions.

Y should be Unit Price

X should be Year

*I have multiple records in a given year, so that's why my unit price is already a AGG function: sum([Net value])/SUM([Order Quantity])