Unexpected Results Using LINESTX for Binary Variable Regression in Power BI

Hi mh20221111 ,

The issue you're encountering is a subtle but common problem with DAX evaluation context, not a bug in LINESTX or an issue with binary variables. Your problem stems from using SELECTCOLUMNS to create the table for LINESTX to iterate over.

When you wrap your table in SELECTCOLUMNS, the subsequent expressions for your X and Y variables ([Cause A] and [Downtime(Hrs)]) are not being evaluated row-by-row over that new, temporary table as you'd expect. Instead, they are evaluated in the outer filter context, which can cause them to be treated as a single, constant value. When the regression algorithm receives nearly constant inputs, it correctly calculates a slope that is virtually zero.

The correct and more direct approach is to apply LINESTX to your original data table. This ensures a proper row context is established, and the function evaluates the X and Y expressions for each individual row of your data. You don't need SELECTCOLUMNS, as LINESTX will only use the columns you specify anyway.

You should use this DAX formula instead, which will give you the expected slope of approximately -6.64:

LINESTX(
    DowntimeData,
    DowntimeData[Downtime(Hrs)],
    DowntimeData[Cause A]
)

It's also helpful to remember what these coefficients mean in your specific case. The intercept (
approx9.05) is the average Downtime(Hrs) for cases where Cause A is 0. The slope (
approx−6.64) represents the difference in average downtime when moving from the Cause A = 0 group to the Cause A = 1 group. So, your analysis correctly shows that cases associated with Cause A have, on average, 6.64 fewer hours of downtime.

Best regards,

VAR RegressionTable =
ADDCOLUMNS(
DowntimeData,
"X", DowntimeData[Cause A],
"Y", DowntimeData[Downtime(Hrs)]
)

RETURN
LINESTX(
RegressionTable,
[Y],
[X]
)

This ensures that [X] and [Y] are evaluated within the row context of the table being passed to LINESTX.