Number of Observations for Robust Statistic


For our corporate reporting we calculate the standard deviation of a list of monthly growth rates.

The common wisdom among my colleagues is that a minimum of 30 such data points is required to calculate a robust statistic and so
we calculate the standard deviation for 3 years.

Can someone please explain what the rationale is for using a minimum of 30 data points?



Not a robit
People like to think when they get around 30 observations, data will approximate the normal distribution, which would also make using measures of central tendency (mean) and dispersion (standard deviation) appropriate.
Well, it depends on the type of data that you're working with. Basically, if you are working with time series obs, then a minimum obs of 24 could be ok, but be careful of the methods, variables size among others. Other data like survey, panel, cross sectional, etc.. are huge sample sizes respectively.