🚀 go-pugleaf

Davis Foulger is adamant, in his example about seatbelts and airbags,
that we would want to examine the effects of each factor
separately even if their interaction was significant, because
some people have only seatbelts and others only airbags.
But these effects are NOT the main effects of an anova.
What you find out if the interaction is significant is that
you cannot talk about the effect of seatbelts or of airbags
^per se^. The effect of each depends on the presence or absence
of the other. So if you have no safety device, installing a
seatbelt will have a different effect to installing it if you
already have an airbag. His example highlights the problem that
expressing the hypothesis tested by the main effects where
there's a significant interaction rarely comes up with a
meaningful statement.

Ronan Conroy                      EMAIL RCONROY@IRLEARN.UCD.IE
Lecturer in biostatistics         Royal College of Surgeons in Ireland
and computing and stuff           Mercer Building, Dublin 2.
                                  (353)-(1)-780200 [fax:780934]
    ---------------------------------
   "Sutor ne ultra crepidam judicaret"