I can not find any paper about, robust statistics where they are looking to outliers not like a outlierts but like that model miss specification.
exp.
i mean that i have sample from N(0,1) ,50 observation and there are value 5 and -4
this 2 are outliers, but TRUE distribution is Caushy and in this case there are not outlieres.
I am looking for paper where autor envisages more possible distribution of errors. And i am looking for some measure of robustness, it is different way like breakdown point.
It is possible to understand me ?
Do you know about something ?
thanks
exp.
i mean that i have sample from N(0,1) ,50 observation and there are value 5 and -4
this 2 are outliers, but TRUE distribution is Caushy and in this case there are not outlieres.
I am looking for paper where autor envisages more possible distribution of errors. And i am looking for some measure of robustness, it is different way like breakdown point.
It is possible to understand me ?
Do you know about something ?
thanks