# GEE linear regression ee.Reducer.linearFit() 'scale'

I am trying to use `ee.Reducer.linearFit()` similar to as used in https://developers.google.com/earth-engine/reducers_regression between 'evi' calculated by me and time.

I observed that absolute values of 'scale' are very high (higher than 1). Since `scale` corresponds to te slope of the line, I expected the value of it to be in between -1 and +1.

What does high values of `scale` indicate? And why aren't the `scale` values a double or float?

• A line closer to the y axis than the x axis has a slope with magnitude greater than 1. Could you add a short example program to your question that demonstrates the calculation you're making? This will help understand whether there's a mistake or how to interpret your particular result. Commented Aug 6, 2019 at 14:14

I don't see anything unusual here. To explore the data values that are going in to the linear fit, I added a layer with the original values

``````Map.addLayer(statsCollection.select('mean').toBands());
``````

and used the Inspector to look at example pixel values (select the Inspector tab next to Console and then click on a point on the map)

``````Pixels
Layer 1: Image (5 bands)
0_mean: 1400.75
1_mean: 1481.75
2_mean: 1865
3_mean: 1138.25
4_mean: 1148.75
Slope of linear fit: Image (2 bands)
slope: -84.75
offset: 171584.9
``````

From this we can see that the slope is not an integer. You were probably looking at the histogram chart's values — a histogram creates "buckets" of some width to count similar values, and in this case chooses integer representative values for those buckets.

The large slope value is not unreasonable, either: just look at the numbers. If we compare the end points it goes from 1400 to 1148, a difference of 242 on the y axis with a difference of 4 on the x axis, or a slope of 242/4 = 60.5 if we "linear fit" only those two points. You could take that number sequence and run the fit using another tool or on paper and I expect you'll get the same answer.

(You didn't ask, but the offset value is large because the data has a large offset in the x-axis: it goes from 2006 to 2010, not around zero! So we can generally expect the offset to be of the magnitude 2008 times the slope — 2008 × 84.75 = 170178, there we go.)