The spending on advertizing likely has diminishing returns. The marginal decrease in satisfaction that would come from doubling the spending is possibly smaller than the baseline decrease caused by the initial expenditure.
There's nothing in their data to support this, and if the effect was there they would have seen it (since they have yearly happiness data). They describe a linear correlation.
They describe a linear correlation because they fitted linear correlation. All kinds of interesting effects could be hidden in the residuals, but they don't show them.
