If the constant is not significant (p>0.05) it means that the constant term is not significantly different from 0. What kind of regression model do you have?

Normally you would not include terms that are non significant. There is some discussion on what to do when it is the constant term, however, I am in favour of regression models that explain the observed data in the simplest way.

If the constant is not significant (p>0.05) it means that the constant term is not significantly different from 0. What kind of regression model do you have?

Normally you would not include terms that are non significant. There is some discussion on what to do when it is the constant term, however, I am in favour of regression models that explain the observed data in the simplest way.

It means that if all predictor(s) have values of zero, that the predicted criterion value will not significantly differ from zero. If the constant is significant however, it means that if the predictor(s) are zero, the predicted criterion value (with predictor[s] of zero) will significantly differ from zero. Remember however, that it is only a statistical interpretation. The zero point is not always reasonable. For example, there is no IQ of zero.

I had a "simple linear regression" model.

My dependent variable was "BMI zscore" and the constant was not significant; however, when I used "BMI" as dependent variable instead of BMI zscore, the constant was significant.

This is probably because BMI zscore is a scaled version of BMI: BMIz=[BMI - mean(BMI)]/stdev(BMI). If your equation is like this: BMIz= b_0 + b_1*x_1 + ..., then b_0 can't be 0 since you would loose the scaling factor stdev(BMI): BMIz = mean(BMI)-BMI + d_1*x1+... And you don't see intercept for BMI probably because the predictors indicate 0 weight when all of them are 0, and of course BMI=0 in this case.

This is probably use to you have already centred your scores around 0 by using Z scores which are zero centred. As Amir states the constant term is there to "un-zero" centre the data and place it around the mean. So, if you are looking at the impact of IQ on BMI, if there was no impact of IQ (and beta=0) a decent regression model should give Y as the mean BMI regardless of IQ. This is due to the constant. If you remove the constant, then if the beta coefficient=0 then Y=0. If the constant is non-significant that means that the mean value of Y is not significantly different from 0. As z scores are zero centred you would hope that this is the case.

So, no need to worry, in fact, this shows that the standardisation process has worked!

Thank you all for the answers.

Hi all, what does it mean when "regression values yield a constant of 6.6 and a coefficient of 0.67"?

So do you include the insignificant intercept in regression equation? Or do you leave it out?

More details here:

https://blog.minitab.com/blog/adventures-in-statistics-2/regression-analysis-how-to-interpret-the-constant-y-intercept