-------------------------------------------------------------------------------
help for apch
-------------------------------------------------------------------------------

Generalized linear models for age-period-cohort effects with hysteresis

Syntax

apch depvar [indepvars] [if exp] [in range] [weight] , [ age(string) period(string) glm_options ]

apch allows all varlists & weights that glm considers.

Description

apch estimates age-period-cohort APC-D (detrended) models and provides detrended (0 sum and 0 slope) parameters of age, period and cohort effects; appropriate constraints offer a unique solution & solve the traditional APC identification problem. We focus on cohort parameters (DCE=detrended cohort effect) that should be tested to be non-zeros across cohort span. Their "hysteresis" (durability) over life course is assessed & tested by a coefficient h (hysteresis) that is 0 in case of linear stability of cohort effect, -1 in case of vanishing of the cohort effect over life span, equal to 1 in case of development from 0 of the cohort effect over life course. Since the cohort parameters and the hysteresis parameter can not be assessed simultaneously, the package proposes an iterative strategy that converges to a single stable solution. The apch provides Raftery's bics able to assess the interest of cohort effect and then of hysteresis effect.

The user provides a dependant variable, age and period, and controls from a mic > rodata series of crossectional surveys. The data must be a complete rectangle (age x period) and the pace between perio > ds must be fix and equal to the distance between age groups. If not, the procedure will > fail.

The procedure delivers : (1) An APC-Detrended model with controls, including zero-sum/zero-slope paramet > ers of age, period & cohort effects. (2) An APC-Hysteresis model with controls, delivering the h-hysteresis paramete > r of linear stability of the DCE. (3) Delta BICs are provided to measure (a) dbic_cohort : the interest of including cohort effects after a AP m > odel and (b) dbic_hyster : the interest of including hysteresis effects to the A > PC model

Options

age(string) and period(string) specify the names of age and period variables. They must be given by the user.

Any glm_options can be valid options for apch.

Many types of model can be handled by apch :

* the user must call "family(bin) link(logit)" for a logit regression

* the user must call "family(poisson) link(log) exposure(pop)", where pop is the population at risk, for a poisson model

* the default "family(gauss) link(id)" will provide a linear regression.

Since apch makes use of complex constraints, this package should be processed on STATA 11.0 or higher.

example

use "http://www.louischauvel.org/apchexb.dta", clear

apch mf2 , age(age) period(year)

References

see http://www.louischauvel.org/apchex.htm for working paper and examples on apch.

see Yang, Y., Fu, W., and Land, K. 2006. apc_ie and apc_cglm STATA packages.

Author

Louis Chauvel Sciences Po Paris During the completion of the v1.0 Louis Chauvel was also invited Pr at Columbia U Dpt of sociology, Knox Hall chauvel@louischauvel.org

Also see

Online: help for glm.