# Is there a calculation in Net Promoter Score (NPS) ?

#### Mikael

##### Quite Involved in Discussions
Hi (so Elsmar is back )

The so called Net Promoter Score (NPS), when I think about it isn't it an Ordinal scale? So you are not really allowed to calculate it?

#### Ajit Basrur

Re: The Net Promoter Score

There is a calculation involved.

Yes, the first step is an ordinal scale from 1 to 10, where all responses with 9 and 10 are treated as Promoters; 7 and 8 considered Passives; and scores 6 and below are Detractors.

After all people have been interviewed with the "one" question, calculate how many promoters and detractors are in the given population, and calculate the NPS by % Promoters minus % Detractors.

#### Miner

##### Forum Moderator
The so called Net Promoter Score (NPS), when I think about it isn't it an Ordinal scale? So you are not really allowed to calculate it?
It all depends on how you plan to use it. While it is true that the average of a set of ordinal data does not meet the strict statistical definition of an average, it can still be useful. I sometimes work with survey data (10 point ordinal scale). If you compare the medians of two groups, they are identical. However, if you look at a histogram of the two groups, they are radically different in shape and spread. An average calculation, does show this difference and is useful for communicating information.

There are numerous nonparametric statistics that are based on ranking the data. Ranking data converts ratio/interval data into ordinal data, whereupon it is used in various calculations (setting a precedent). Therefore, when you use ordinal data, you are essential using a nonparametric approach.

#### Mikael

##### Quite Involved in Discussions
Sorry for a late reply here, I guess I lost the thread.

I cant recall why I put the question in the headline like that, of course there is a calculation (I had been using it at work), but thx Ajit anyway. What I wanted to discuss is of course whether it is nonsense to do these calculations.

I read somewhere that when you use frequencies over the whole spectrum, it should be okay, because then you have an interval? There was twice as many replying "..." . Or is it still a cover?

Miner, thx, though I am not sure about the nonparametric approach, just calling it that does not legitimate it?