The making of a chimera

Borne out of confusion between prediction and explanatory modelling

Chimeras are rare indeed. Nevertheless they do come out often. Nature is against it. However occasional mating in the dark wilderness occurs. It may happen by mistake, confusion or helplessness. Sometimes you are in the wild, alone or with a few of your own, amidst a clan of another similar species. Very rarely, they will mate with you by coercion.You yield under your utter helplessness and quest for survival. Your people are unable to protect you. And a useless and grotesque chimera is born.

In epidemiology, most of the research is related to descriptive or associational questions or an attempt at explanation. In to this came, the damsel the prediction modeling. Though regression modeling is used in the analysis of these, difference are aplenty at the levels of design, planning, selection of variables, specification of models and in the intricacies in the fine art and science of modeling. In associational and related studies, we are more interested on the right side of the regression equation, and how the expectation of outcome on the left side of the equation is related to the factors on the right side of the equation, how a change in one or more of the factors on the right side reflects on the observed outcome ie the Y on the left side of the equation. Our interest is in the factors on the right side and how a change in one or more factors on the right side of the equation reflects as a change in the observed outcome on the left side . In this scenario, exposure variable is important,confounders and multicollinearity more relevant. Statistical association of each individual factor with outcome may be pertinent. One of the important offshoots is the potential intervention or modificaton that can be applied to the modifiable factors on the right side of equation, with many concomitant potential policy changes and implications. If the factors on the right side of the regression equation explain the expectation on the left side of the regression equation, under a well constructed study with correct hypothesis and with a theory informed explanatory variables selected and with no sampling bias or measurement errors ,all confounders taken care of , along with correct interactions and forms specified and with minimal or no collinearity and many other subtlities , then there is a great role for the modifiable factors on the right side of the equation from a public health angle or from a policy makers perspective for the intervention of the potential treatable variables from a high risk or a population lens .

However the picture is different in prediction modeling, entirely different. When you bring in the underlying concepts, philosophy and nuances of regression modeling relevant in associational and explanatory question to prediction modeling, the end result may be a perfect tailor-made model, a perfect fit for the data at hand , but a sure recipe for failure in new data.

In prediction modeling, what one aims at is making a ready-made shirt that can be worn by a lot of future or present potential customers who fall in the size range the ready made shirt is made for.

To continue……………………….

Related posts:

Do human beings exist just to go to school, look for a job, work, pay bills, and die? Or is there a grander scheme of things? Every individual who is a seeker of truth will have their own individual…

Aunque la palabra hacker se ha utilizado en el pasado con connotación negativa, hoy, la realidad es otra.

The world is becoming more connected and involved with online activity thanks to the ability to access the internet through our multiple devices and communication infrastructures. But with the…