I am not asking about negative values. I understand that is possible. But my intercept is bigger than my max value. Is something wrong with my data or model? I can't find anything on range of values for intercepts. Conceptually though it seems it would need to be within the absolute value range of the DV.
Thanks,
Of course it's possible, because the intercept is what happens when x==0, which might be quite far away from where the actual observations lie along the x-axis. It's easy to create an example. For example, here is the `airquality` dataset that comes with the base installation of R. In this graph, the solid red line shows the regression line, the X represents the intercept, and the dashed red line shows that the intercept is higher than any data point.
Hi,
Could you provide us an scatterplot of the dataset? I think that this scenario is perfectly feasible.
Of course it's possible, because the intercept is what happens when x==0, which might be quite far away from where the actual observations lie along the x-axis. It's easy to create an example. For example, here is the `airquality` dataset that comes with the base installation of R. In this graph, the solid red line shows the regression line, the X represents the intercept, and the dashed red line shows that the intercept is higher than any data point.
But what is the range of wind ?? In my example I have GPA. GPA is not ever higher than 4.0. However, my intercept is about 7. This does not even make sense. If the intercept is when all else is equal to zero - my white males would not have a GPA of 7.
I understand otherwise that it can move, but how can it be outside of the range of the DV?
Thanks,
Like I said above, it depends on what your predictors are. Imagine one of your predictors is some number that's up in the tens of thousands. Here's some fake data where the predictor is how far away from campus the student lives, in feet. All the data are up around 16,000. So you can see very clearly if you follow that regression line all the way left to x=0 feet, then it's going to be well above 4.0 GPA. (In fact the intercept for this particular model on this particular dataset is about 13.)
The range of the outcome variable doesn't matter; an ordinary regression model doesn't know it has limits. it just fits a regression line to some data and lets that line go on forever in each direction. If the data are clustered super far out along the x-axis somewhere, then the intercept will necessarily be super high or low along the y-axis, assuming the line has a nonzero slope.
I suggest you just draw a graph of your data and the associated regression line, as Raúl explained, and then I think this will be clearer.
That doesn't mean the intercept "did not make sense". It makes perfect sense for what it is --- the height of the regression line when all predictors are at zero. The issue is that, like I said above, if some variables' data are all very far away from zero then this intercept might be well outside the range of the data. Centering that variable makes 0 be right in the middle of the data, and thus makes the intercept also be in the middle of the data. But none of this means the original intercept had any sort of problem; it just means the original intercept couldn't be interpreted as a mean or anything like that.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression