The effect of grouping on regression analysis

2017-06-06T01:01:37Z (GMT) by Beaumont, Nicholas
Social science data is often collected in grouped form, respondents indicating that their ages and incomes are in ranges such as 30 -39 years and $40,000 to $49,999 respectively. Although grouping biases and makes less precise estimates of population statistics, social scientists appear to have ignored its effects. Its effects are usually mild but very common and occasionally severe (it can exacerbate the effects of multicolinearity). In many cases ignoring grouping makes tests of hypotheses weaker than they need be. In some cases, it is easy to modify statistical formulae to partially remove bias and recover some otherwise lost statistical power. The paper's assertions are based on theory verified by simulations.



In Copyright