Miscellaneous Flashcards
Null and alternative hypotheses for f-test?
Null: Coefficient on all variables on the model are jointly zero
Alternative: Coeffiicient on all variables in the model are NOT jointly zero
Hausman test
evaluates the consistency of an estimator when compared to an alternative, less efficient estimator which is already known to be consistent
(ie used in context of comparing RE and FE)
Null of hausman when using IV?
Null is that the variable is exogenous, so you don’t need to use IV
Reject=need to use IV
Null of hausman in RE and FE?
Null is that RE and FE estimates are the same
Reject=need to use FE
What does it mean to have serial correlation in a DD model?
we have repeat observations of the same group; there
is likely to be correlation in their error terms over time. This violates Gauss-markov assumption that error terms are iid
If there is serial correlation in a DD model, what happens to standard errors?
They are too small–not accounting for the fact that an individual’s errors are correlated over time
how can you account for serial correlation in DD?
cluster standard errors and also try to get larger sample
How is sample size related to severity of serial correlation in a DD model?
Most severe=few individual units that are observed over a longer time period
Less severe=many units observed over only a few time periods.
selection bias v sample selection bias?
selection bias-individuals in T and C groups are different, and their potential outcomes are different
sample selection bias-probability that an individual is inn the sample is related to the outcome (ie differential atrition)
why would you take the log of an IV?
taking the log of a variable condenses the range of a variable and makes the interpretation of the dependent variable in percentages, not units.
It’s useful when you have a variable that follows an exponential distribution with widely disparate levels
What is the pseudo r-sqaured?
A measure of goodness of fit for a logistic regression
It is the ratio of the log likelihood of the full model to the log likelihood of the model with only the interecept (the restricted model)
what does the t critical value indicate?
how many standard deviations away from 0 a certain percentage of the normal or t distribution lies
what are you doing when you multiply tcrit* se?
Convert t distribution to one reflective of current mean and se
what does the product of the tcritical value and se indicate?
how many se’s away from the mean 95% of the distribution would fall
how is t statistic chosen?
so that the area between -t and t contain a desired portion (often 95%) of the distribution
