describing data about 'rates'

[BillRoberts]

Hello Data Cubists

You frequently come across statistical data expressed as a rate - crimes per 1000 population, rate of heart disease per 100,000 people in the age range 0-75 etc.

When representing this in RDF Data Cube form, I have typically defined what is being measured either in the definition of the measure property, or in some cases in a unitMeasure attribute.  However I'd like to be more specific about what the numerator and denominator are in the ratio that constitutes the value of the measure.  

The numerator is usually the count of some quantity that we can define with qb:concept or similar.  The denominator is sometimes straightforward but often it is more complicated (eg people within a particular age range, or population estimated at a specific time or using some particular method).  And if you want to know if and how a dataset of rates can be aggregated or averaged, you need to know what to multiply the rates by in order to do the aggregation.  

Has anyone come across or created a systematic way of doing this?  I imagine you could relate an observation of a rate to an observation of the corresponding numerator and denominator from different data cubes.  Is there a predicate in SDMX or RDF Data Cube that you would recommend using for this?

Thanks

Bill

[Dave Reynolds]

Hi Bill,

There's no standard approach to this specifically within Data Cube.

We did log it is an issue for the W3C GLD work [1] but didn't have time
to tackle it properly so it is current "POSTPONED". As part of the input
to this Arofan did mention [2] that the SDMX community has done
considerable work representing the computation process for how values
are derived, which is a more general capability that we imagined for
Data Cube.

There is some capability along these lines (well aggregation at least)
in the QB4OLAP work [3] but that doesn't go as far as you need here.

You could imagine using slices of the base cube (e.g. population in your
case) to define the denominator and then have some sort of DENOM
relationship between the measure in the derived cube and the relevant
slice. I've done that for some experimental work in the past but have
no detailed standards proposal for it.

Dave

[1] http://www.w3.org/2011/gld/track/issues/30
[2]
https://groups.google.com/forum/#!msg/publishing-statistical-data/3I1-Ix1Hk14/HVYzjD0C_bMJ
[3] http://publishing-multidimensional-data.googlecode.com/git/index.html

> --
> You received this message because you are subscribed to the Google
> Groups "Publishing Statistical Data" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to publishing-statisti...@googlegroups.com
> <mailto:publishing-statisti...@googlegroups.com>.
> For more options, visit https://groups.google.com/d/optout.

[Richard Cyganiak]

Hi Bill,

I can’t offer relevant practical insight, but a general comment: Perhaps these cases can be modelled as mathematical relationships between the measure properties (e.g., the measure:crimeRate property is defined as measure:numberOfCrimes divided by measure:population). This might be better than treating them as mathematical relationships between observations. I say this because relating the properties would mean that you take the problem out of Data Cube land and into general “modelling things with RDF” land, so you would be able to use a generic, non-data-cube-specific solution for expressing the relationships. (Not that I am aware of such a solution that would be useful here.)

Best,
Richard

> --
> You received this message because you are subscribed to the Google Groups "Publishing Statistical Data" group.

> To unsubscribe from this group and stop receiving emails from it, send an email to publishing-statisti...@googlegroups.com.