Analysis using 'lurking' variables

Variables are only called 'lurking' variables if they are either unrecorded or are unused when analysing the data. In order to explain the resulting problems, we have used examples in which the values of the lurking variables were available. To show the potential for reaching the wrong conclusions, we presented two analyses of most data sets:

  1. A simple naive analysis ignoring the lurking variable, and
  2. A more complete analysis using this variable (which should therefore no longer be called 'lurking')

Although the simple analysis can result in wrong conclusions, a full analysis using the 'lurking' variable is always more complex.

Do not ignore lurking variables to simplify the analysis — you could reach the wrong conclusions from the data.

We now present a few more examples where lurking variables can lead to the wrong conclusion from data.

Ice cream and murders

Weekly ice cream consumption and murder rates are highly correlated. Does ice cream incite murder or does murder increase the demand for ice cream? Neither — the relationship that is observed is the result of a lurking variable, the weather. Both ice cream consumption and murder rates increase in the summer.

If average temperature was recorded each week, it would be possible to examine how the murder rate is modelled by both temperature and demand for ice cream. (This would probably show that only temperature is an effective predictor of the murder rate!) However this analysis is rather more complex than a simple scatterplot of murder rate against ice cream demand.

Damage from house fires

In an analysis of damage caused by house fires, it was found that the more firefighters that are sent to a fire, the greater the damage. Are the firefighters causing extra damage? The lurking variable here is the size of the fire — large fires both cause more damage and get attended by more firefighters.

Again, a full analysis taking account of fire size is more complex but is essential to avoid reaching the wrong conclusion.