Discrete dependent variable models:

Standard linear regression models are usually adequate for analyzing  problems where the behavior of the economic agents may be approximated by  continuous variables. However, whenever  a choice has to be made between  two or more alternatives the continuous approximation  is  often  not  a  good  one. Examples of choice situations include:  

1  )  For  individuals: Whether to work or not; whether to attend college; whether to marry;  choice  of occupation;  number  of  children; whether to  buy  a  house; purchase of consumer durable goods; what brand of a consumer durable good to purchase; whether to migrate, and  if  so where; whether  to go on  vacation and where.

2)  For'  firms:  Whether  to  build  a  plant,  and  if  so,  at what  location;  what commodities to produce; whether to shut down, merge or acquire other firms; whether to go public or private; whether to export or not; whether to spend on research and development and how much; whether to accept union demands or take a strike. In  all the above situations, we can construct models that  link  the decision taken (or outcome)  to a set  of  factors, which is  similar  to what regression analysis  does. However, as we shall see below none of these situations can be effectively analyzed using standard regression analysis.

In  all  the above cases the variables generated  by  the agent's  behavior are discrete valued. To analyze  and understand  this  behavior we  need  to  build  models which approximate the actual data generation process. As  is clear from the examples given above  the discrete variables  generated  may  be binary  (having only  two possible values)  or  multinomial  (having  greater  than  two  but  a finite  number  of distinct values. Several,  different methods of  modeling  and  analyzing  discrete data are
available  today.