Average of fractions: Weighted sum of numerators over weighted sum of denominators?

1.5k Views Asked by At

I am dealing with data corresponding to demands for $N$ products. The data consists of demand $d$ of each product that vary by month, and the quantity $m$ of that demand that is met. I also have different $d$ and $m$ data by store. Generalizing, $d$ and $m$ vary by different "independent" variables (the product, the month, the store, and others). I want an intuitive average of $m/d$, and the average will take place over different subsets of the data by freezing certain independent variables. In case it helps, this is the SQL group-by function, with counterpart idioms in Matlab and probably other analysis environments.

I want to weight the averaging by a scarcity weighting, which is basically one over the quantity of the particular product that is ordered each month in central stores. Since there are $N$ products, each $d$ and $m$ value will be weighted by one of $N$ different scarcity weights $w_i$. In the simplest possible form, an average on two values for $d$ and $m$ would be:

$$ \frac{w_1 m_1 + w_2 m_2}{w_1 d_1 + w_2 d_2} $$

I thought that it would be simple to show the equivalence with the usual weighted average:

$$ \frac{ w_1 m_1 / d_1 + w_2 m_2 / d_2 }{ w_1 + w_2 } $$

It turns out that they are not equivalent. It's certainly not a geometric average. It also doesn't seem to correspond to a weighted harmonic average. Yet it seems so intuitive. Can it be related in some intuitive way to a classically defined average?

The supply/demand problem that I used to provide context to the average is just an analogy, as the actual problem is different. So while I understand why respondents may question the nature of the problem, it's the actualy math that I'm hoping to rationalize in terms of classical definitions of averages.

1

There are 1 best solutions below

0
On

Thanks to Rahul for the answer. The average is known as a weighted mediant.