Upgrade to remove ads
ECO 418 Exam 2
Terms in this set (33)
multiple entities and one time period
multiple entities and multiple time periods
multiple time periods and one entity
1% t-stat significance
t-stat >2.58 or 99% confident
5% t-stat significance
t-stat>1.96 or 95% confident
t-stat > 1.64 or 90% confident
f-test for joint significance
Null hypothesis beta one = 0 and beta two =0
alt hypo beta does not = 0 and beta two does not = 0
when to transform data into log
1. when a variable is a positive dollar amount
2. variables that take large integer values
3. don't use logs if a variable takes on zero or neg
4. do not log a variable if it is already expressed as a percent
4. consider when it make sense. dont use on index variable.
product of two variables.
one dummy and one continuous; 3 cases
different int. same slope: non interaction. one fold
different int, diff slope: interaction. two fold
same int, diff slope: interaction. two fold. no individual effect
how can you apply your results to other entities and time periods
a study is externally valid if the statistical inferences and conclusions obtained from the empirical results can be generalized more broadly.
good regression equation that doesn't violate any assumptions
a study is internally valid if the statistical inferences about casual effects are correct for the population being studied. valid if you have done a good job with the data you used. coefficient estimates are unbiased and consistent
threats to external validity
differences in population: can a study of mice be applied to people?
difference in setting: time and place
threats to internal validity
omitted variable bias: occurs when 1. a determinant of Y is omitted from the regression and 2 that variable is correlated with the explanatory variable of interest
2. misspecification of the functional form: not use quadratic to identify nonlinear and incorrect log.
3. measurement errors
4. same selection bias. when observations are not randomly selected
5. simultaneous causality: x explains y and y also explains x.
balanced panel data
data for each entity is available for all time periods
at least one entity is missing data for at least one time periods. problematic cause it skews data.
signficance of an event
setting up before and after dummy variable. before is 0 after is 1. only use when before and after is key in the results
fixed effects: used to control for variation across entities or across time periods. this is done by creating dummy variables for each entity or time period. intended to capture unexplained variation across entities/time periods. the dummy variable absorb the influences of all omitted variables that differ from one entity to the next.
keep time dummies as regressors even if they are not significant
time fixed effect
always include when doing panel data
dont include only if 1. you include a before and after time dummy as a regressor
2. the number of time periods is much larger than the number of entities.
occurs when the regression residuals in a time period t are correlated with residuals in time period t-1. this often causes standard errors for a regressor to be biased or more specifically smaller.
why OLS cant be effective used to estimate equations with binary dependent variable
binary means y= 0 or 1. OLS assumes an linear relationship. and linear probability model. there is no upper and lower boundary.
marginal effects calculates the marginal effects (or elasticities) at the mean of the independent variables. ME provides a good approximation of how a one unit change in x impacts the probability that Y=1
MFX are identical for both probit and logit.
use a tobit regression when the dependent variable is censored or truncated on the right, left, or both. there are a large number of observations at the cutoff on either end.
ordered probit and ordered logit
this regression model is used when the dependent variable is NOT a binary dummy variable, but rather a series of naturally sequential choices.
how do you interpret the results of an ordered logit or probit
the probability of transitioning into the next group
can you specify mfx a
you can use mfx if the number of right hand side regressors is greater than the number of categories in the dependent variable
use a multinomial logit if the dependent variable is not a binary dummy variable but rather a series of choices that are not naturally sequential
utilized when the dependent variable is 1. integer value. AND 2. small range
2 probs function for count data
1. poisson: use when its variance is less than its mean. not over-dispersed. coefficients are only interpreted in terms of direction and significance
2. negative binomial: use when its variance is larger than its mean. over-dispersed. interpret same as poisson
if one or more regressor is correlated with the error term, then x is an endogenous variable
if the regressors are not correlated with the error term, then x is an exogenous variable
three reasons why regressor is correlated to error. corr(x) not = 0
1. Omitted variable bias
2. measurement error in the variables
3. simuleneity. aka endogeneity.
two criteria that a variable must meet to be a suitable IV
1. instrument relevance: the instrument(exogenous) and the regressor are correlated.
2. instrument exogenetiy: the instrumental variable is not correlated with Y
THIS SET IS OFTEN IN FOLDERS WITH...
ECO418 Chapter 5
YOU MIGHT ALSO LIKE...
Research Methods Concept Quiz
Stats Exam 2
Social Research Final Exam Review 2
Research @ TSS SU2016
OTHER SETS BY THIS CREATOR
Business Stats Exam II