teaching microeconometrics using at warsaw school of economics marcin owczarczukmonika książek
TRANSCRIPT
Teaching Microeconometricsusing
at Warsaw School of Economics
Marcin Owczarczuk Monika Książek
Agenda
• What is microeconometrics
• Microeconometrics – the lecture
• How do we teach:• Ordinal outcome models• Count outcome models• Limited outcome models
Microeconometrics
• Microdata• Individuals• Households• Companies
• Microeconometrics = econometrics for microdata
• Fields of application:• Marketing• Finance• Social science
Microeconometrics – the lecture
• 15 lectures (2h each)
• Theory + applications
• Applications on publicly avaiable datasets
• Calculations in STATA
• Maximum likelihood
• Binary, multinomial, ordinal, count, limited dependent variables
• Cross-sectional data only
Ordinal outcome models
Data
• European Social Survey, vawe 3, Poland• Ordinal dependent variable (ocdoch):
Which of the descriptions on this card comes closest to how you feel about your household’s income nowadays?1 Living comfortably on present income2 Coping on present income 3 Finding it difficult on present income 4 Finding it very difficult on present income
• Independent variables:• Continous AGE (wiek)
• Binary CHILDREN (dzieci)
• Nominal (3 categories) PROFESSION (zawód: kierownicy, pracownicy)
OLOGIT, OPROBIT, GOLOGIT
Significance testing:• Single variable• Variable set• Whole model
Parallel regressions assumption testing
• Brant • Wolfe & Gould
• LR ologit vs gologit
Assumption holds standard model is OK
Model quality assessment• Model fit
• Predictive capacities
predict prob1, outcome(1)
Parameters interpretation• Compensating effect• Marginal effect
• Odds ratio
Count outcome models
Data
• CBOS survey: Living conditions of Polish people – problems and strategy
• Dependent variable: number of small children (up to 6 year old) in a young family (20-35 year old)
0.2
.4.6
.8D
ensi
ty
0 2 4 6V344
Poisson regression
Negative binomial regression(allows for overdispersion)....
No overdispersion Poisson model is OK
Zero inflated (Poisson) model
ZIP fits better than standard Poisson model
(Binary logit model: P(Y=0))
(Poisson model)
Limited outcome models
Data
• PVA (US not-for-profit organisation) which rises funds by direct mailings
• Donors differ in amounts and frequencies of gifts
• Explanatory variables• history of previous mailings• characteristics of the donor’s neighbourhood
Tobit regression
Target_d – amount given in last mailing (many zeros)
Truncated regression
Target_d – amount given in last mailing (no zero observations)
Sample selection, maximum likelihood
Srednia_odleglosc – average distance (in days) between gifts;
sredni_datek – average amountselekcja =1 if more than 6 gifts were given
Positive correlation – who gives more, gives less frequently
Significant correlation
Sample selection, two step
Inverse Mills ratio
Coming soon
Coming soonSeptember 2010
September 2010