[OSM-talk] Survey coming

[Christoph Hormann]

On Sunday 10 January 2021, Allan Mustard wrote:
> The respondents constitute a sample of the target population. Since
> we don't really know what the population looks like (we don't even
> know the breakdown by gender, do we?) we are aiming for as large a
> sample as possible.  With a large enough sample, the law of big
> numbers [1] takes over.

No, sorry, that is mathematically incorrect.  A larger sample size will 
not reduce any *systematic* bias you have in sample selection, it will 
just reduce the noise resulting from *randomness* in sampling.

That does not mean a large sample size is not useful, in particular if 
you want to weigh the resposes differently to compensate for a known 
bias in selection but a large sample size is not in any way a 
substitute for ensuring an unbiased sampling.

Even without knowing your target population composition precisely 
(though we do know some things with some degree of certainty - for 
example geographic distribution of active mappers) you can do a lot to 
minimize bias in survey design.

-- 
Christoph Hormann
http://www.imagico.de/