ECON 1 Lecture Notes - Lecture 12: Multivariate Analysis, External Validity, Golden Rule
MEETING 3: METHODS
Correlation vs causation
For sociology correlation is enough, in economics we need causation
𝑦𝑖 = 𝜎𝐷𝑖 + 𝜉
participation to honors program
D=1→H student
D=0→not part of the program
We are interested in the impact of additional resources
Y can be labor market data in later time
𝜎→this is likely to be a correlation and not causation, there might be additional variables that
influence the impact called Endogeneity problem
• Reversed causality
• Omitted variable bias
• Unobserved heterogeneity
Omitted Variable Bias: in this case it can be skills or previous grades; previous grades influence
both the treatment (being part in the honors program) and the outcome→treatment is correlated
with the error term
1 solution is to control for variation by adding more variables (multivariate analysis) regarding the
observed characteristics→you can’t control for unobserved variables
𝑦𝑖 = 𝜎𝐷𝑖 + 𝛽𝑥𝑖 + 𝜉
xi is grades
SO→randomization of the treatment, randomly assign D, in this case half of the population does
the honours program, data=cross section data
OVB is not an issue because by randomly allocating the treatment→we make 2 populations that
are equivalent so we control for the variations that are given for example by x
So, you expect 𝛽 = 0 if you have perfect randomization and then you would have a causal
interpretation
Also, reverse causality is not an issue here because any variable that is correlated with the
outcome should have a coefficient of zero
Like first year GPA should not influence the participation in honors program
Golden rule standard→control, you randomize the treatment
Problems:
1. Ethical concerns for the ones that are not selected in the program
2. External validity (what can we learn from the study→is it generalizable?); internal validity
(how valid is the study); experiments in artificial settings do not always yield to accurate
results in real life