3
$\begingroup$

I am reading a research paper which models a regression model where the returns are regressed on the number of ad exposures.

the equation looks something like this:

$Returns = beta_1*nExp + beta_2*nExp^2$

nExp: The number of times the user looks at the ad.

Returns: The revenue from the user

The result is that: $beta_1$ is positive (and statistically significant), which means that the ad exposures has a $+ve$ relation with the revenue.

However, $beta_2$ is negative. And the research paper says:

term for the square of the number of exposures, which has a negative coefficient suggesting diminishing returns to ad exposure.

What does it really mean? What does it mean when the term's coefficient is $+ve$ but the coefficient of it's squared term in the equation is $-ve$?

$\endgroup$
8
$\begingroup$

In this context probably the plain English way to put it is that 'returns increase with additional exposure to the ad, but there is a tapering effect at the upper end of exposures. Looking at this picture of a parabola here (i.e a graph of y= -ax^2+bx+c):enter image description here

Most likely your returns data are between 0 and the peak so you don't actually see a decline, just a reduction in the rate of increase.

$\endgroup$
  • 1
    $\begingroup$ +1, spot on. The fitted response to Exp is parabola-shaped (because you have a squared term, but no higher powers), and the parabola opens downwards (because the estimated leading coefficient $\beta_2$ is negative). Depending on where you are on the $x$ = Exp axis, you see an increasing response with a diminishing positive slope, or a decreasing response with a negative slope that increases in absolute value. I recommend plotting the estimated equation over the range of Exp you have actually observed. $\endgroup$ – S. Kolassa - Reinstate Monica May 27 '16 at 9:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.