ECON 424

13. Which one is correct about correcting issues of heteroscedasticity and autocorrelation? Check all that apply. (One correct answer.) a. Sometimes using functional forms such as logarithmic functions corrects the issue. b. They cannot be corrected if we do not know the exact mathematical form of the issue. c. Including missing explanatory variables always corrects the issue. d. They should always be corrected using WLS.

a

15. Which one is correct for tests of detecting autocorrelation? (Two correct answers.) a. Durbin's alternative test statistic has a chi-squared distribution. b. If errors are autocorrelated, Durbin's alternative test statistic is close to zero. c. If errors are autocorrelated, Durbin-Watson test statistic is 2. d. Durbin-Watson test is a test of detecting autocorrelation with a test statistic the takes a value between zero and four.

ad

29. Which one is correct about the three R-squared measures calculated in the fixed-effects and random-effects regressions of a panel of students' scores and their study time? Check all that apply. (Two correct answers.) a. Between R-squared is based on ignoring the variations in individual scores over time and works with the average score of each individual over time. b. Overall R-squared is the average of within and between R-squared. c. Overall R-squared is the sum of within and between R-squared. d. Within R-squared describes how much of the variations in individual scores can be explained by the variation in their study time, regardless of their differences with other students.

ad

6. Which one of the following qualitative variables can be used as a dependent variable? Check all that apply. (Two correct answers.) a. Union membership (member/non-member) b. Race (white/nonwhite) c. Gender (male/female) d. Homeownership (owner/non-owner)

ad

25. Which one is correct about the error correction model. Check all that apply. (Two correct answers). a. It discovers the short-term relationship after adjusting for the long-term relationship. b. It requires that the residuals of the long-term relationship be a random walk. c. It contains the errors from the long-term regression to adjust for a possible long-term relationship. d. It always has a significant coefficient for the residuals of the long-term regression.

ac

31. What is the problem of counterfactual in program evaluation? Check all that apply. (Two correct answers.) a. The average difference between outcomes in the control group and treatment group is not the causal effect of treatment. b. Control group is not available in most programs. c. At any point in time, an individual is only in the treatment group or control group. d. Members of the control and treatment groups have similar features that make them inseparable.

ac

16. Which one is correct about time series data? Check all that apply. (Two correct answers.) a. It is likely that time plays the role of an independent explanatory variable. b. It is more prone to the violation of independence assumption. c. It is a series of observations taken at one point in time. d. Prediction and forecasting are not as important in time series data as in cross-section.

ab

24. Which one is correct about two cointegrated I(1) time series? Check all that apply. (Two correct answers.) a. A linear combination of them is stationary. b. We must use Dickey-Fuller test to detect the stationarity of residuals of the regression of the two variables. c. Regression of the first difference of two conintgrated time series detects genuine long term relationship, if there is any. d. The difference between the two variables is stationary.

ab

34. Which one is correct about the dif-in-dif method? Check all that apply. (Two correct answers.) a. The differences between treatment and control group before the introduction of the program is used to remove the bias. b. Its accuracy hinges on the equality of secular trends in the control and treatment groups over time. c. Regression of the outcome variable on treatment dummy provides the causal effect of the treatment. d. The coefficient of the time dummy variable captures the secular change over time in the treatment group.

ab

11. Which one is correct about the effect of heteroscedasticity? Check all that apply. (Three correct answers.) a. Under heteroscedasticity, OLS estimators are no longer BLUE. b. Under heteroscedasticity, variances of OLS estimators are no longer valid. c. Under heteroscedasticity, hypothesis testing may result in an incorrect decision. d. Under heteroscedasticity, OLS estimators are no longer unbiased.

abc

18. Which one is correct about AFC and PACF? Check all that apply. (Three correct answers.) a. ACF is a series of coefficients of correlation between the variable at time t and its lagged values. b. We use AFC and PACF to identify the type of the model and its order, i.e., number of lagged valued. c. PACF is a direct coefficient of correlation between the variable at time t and its lagged values, after taking the effect of other factors into consideration. d. The first bar in AFC is always bigger than the first bar in PAFC.

abc

30. Which one is correct about fixed-effects models and random-effects models? Check all that apply. (Three correct answers.) a. Random-effects models assume that individual characteristics are completely random factors. b. Fixed-effects models assume that there is a correlation between individual characteristics and idiosyncratic error terms. c. Fixed-effect estimators are better than random-effects estimators if there is indeed a correlation between individual characteristics and idiosyncratic error terms. d. Fixed-effects estimators are always better than random-effects estimators because they account for the endogeneity problem.

abc

33. Which one is correct about issues we face in RCTs? Check all that apply. (Three correct answers.) a. It is usually very costly and time consuming. b. Individuals may behave different from their normal-life behavior simply because they participate in a program. c. Individuals make choices, and these choice may mix up control and treatment groups. d. Designed treatment and control groups may not be comparable.

abc

23. Which one is correct about the relationship between two non-stationary of degree one variables? Check all that apply. (One correct answer.) a. They are correlated only because of their nonstationary nature, not because of a real relationship. b. They may or may not have long-term and/or short-term relationship. c. If they are cointegrated, they have a long-term relationship, but not a short-term relationship. d. They are necessarily correlated in the long term.

b

32. Which one of the following methods relies heavily on observable variables to define counterfactuals? Check all that apply. (One correct answers.) a. Regression discontinuity method. b. Matching methods. c. Difference in difference method. d. Instrumental variables method.

b

9. Which one is correct about odd-ratio? Check all that apply. (Two correct answers.) a. It is a linear function of explanatory variables. b. It is the probability of dependent variable being one relative to the dependent variable being zero. c. Its log (log odd function) is a linear function of explanatory variables. d. It is always between zero and one.

bc

2. Which one is correct about dummy variables? Check all that apply. (Two correct answers.) a. When a categorical variable with no order has many categories, such as states of the United States, we can assign a number to each category and use it as a single numerical variable. b. Categorical variables with natural orders, such as customers' evaluation of a commodity, among categories can be converted into regular variables OR a set of dummy variables. c. Categorical variables with no natural orders, such as types of car, among categories MUST be converted into a set of dummy variables. d. Numerical variables such as income cannot be converted into a set of dummy variables.

bc

20. Which one is correct about forecasting in time series data? Check all that apply. (Two correct answers.) a. Forecasting the future is impossible, so we rarely discuss forecasting in time series data. b. Forecasting refers to "out-of-sample" or simply future values of the variable of interest. c. Forecasting is one of the main goals of time series analysis. d. Forecasting in time series is the same as prediction in cross section-data.

bc

27. Which one is correct about three types of variations in a panel data of students' scores? Check all that apply. (Two correct answers.) a. Within variation is always larger than between variation. b. If each student always gets the same score, within variation is zero. c. If instructors believes in perfect equality of students in scores, between variation is zero. d. If each student always gets the same score, overall variation is zero.

bc

4. Which one is correct about using dummy variables in regressions? Check all that apply. (Two correct answers.) a. If a qualitative variable has k categories, we should define k dummy variables to cover all information. b. Adding multiplication of dummy variables, similar to adding dummy variables, only SHIFTS up or down the fitted lines between the variable of interest and numerical variables. c. In dealing with gender, adding a dummy for male and another dummy for female is not accepted, because of perfect collinearity. d. We cannot add multiplication of two dummy variables along with the two dummies in a regression because of perfect collinearity.

bc

5. Which one is correct about using multiplication of dummy variables and numerical variables in regressions? Check all that apply. (Two correct answers.) a. The sign of interaction between a dummy and a numerical variable is the multiplication of the sings of the dummy variable and the sign of the numerical variable. b. It is possible that both dummy variable and the numerical variable are statistically significant, but the interaction is completely insignificant. c. We use the interaction between dummy variables and numerical variables to analyze the different effects of numerical variables on variables of interest for different groups in the dummy variable. d. Adding multiplications of dummy variables and numerical variables is always more explanatory than adding dummy and numerical variables separately.

bc

1. Which one is a qualitative variable and should be turned into a dummy variable? Check all that apply. (Three correct answers.) a. The number of children a family has. b. Income group a person is in, such as low, middle, and high income. c. Commuting-to-work methods. d. Living in one of 50 states of the United States.

bcd

35. Which one is correct about coefficients in the dif-in-dif method? Check all that apply. (Three correct answers.) a. The coefficient of treatment variable captures the difference between the control and treatment group after treatment. b. The constant term in the regression shows the average value of the outcome variable for the control group before treatment. c. The coefficient of treatment variable captures the difference between the treatment and control group before treatment. d. The coefficient of interaction between time dummy and treatment dummy is the dif-in-dif estimator and shows the causal effect of program.

bcd

8. Which one is correct about logit model? Check all that apply. (Three correct answers.) a. The estimated coefficients show the effect of an extra unit of explanatory variables on the probability of dependent variable is one. b. Regular R-squared is not valid for the model. c. It cannot be estimated using the OLS method; we use the Maximum Likelihood method. d. Its predicted probability is always in the range of zero and one.

bcd

10. Which one is correct about pseudo-R-squared and its components? Check all that apply. (Two correct answers.) a. Log-likelihood of the model and log-likelihood of the intercept are both positive. b. If the model has no explanatory power, its pseudo-R-squared is zero. c. Log-likelihood of the model is always bigger, in absolute term, than log-likelihood of intercept (a model with no explanatory variables). d. Pseudo R-squared is always between zero and one.

bd

12. Which one is correct about autocorrelation in errors? Check all that apply. (Two correct answers.) a. Autocorrelation is more likely to happen in cross-section data than in time series. b. With autocorrelated errors, OLS estimators are no longer BLUE. c. With autocorrelated errors, OLS estimators are biased. d. With autocorrelated errors, regular test statistics calculated using OLS are no longer valid.

bd

17. Which one is correct about stationarity? Check all that apply. (Two correct answers.) a. Stationarity is the idea that past and future are structurally different. b. Covariance stationarity is a less restrictive version of stationarity. c. In stationary data, the covariance between observations a constant number for all lags d. In stationary data, the average and the variance of observations remain constant over time.

bd

19. Which one is correct about autoregressive of order one? Check all that apply. (Two correct answers.) a. The coefficient of lagged value can be positive or negative, but its absolute value is more than one. b. The effects of shocks gradually disappear over time. c. Its autocorrelation function has only one spike at the first lag. d. Its partial autocorrelation function has only one spike at the first lag.

bd

21. Which one is correct about a random walk process? Check all that apply. (Two correct answers.) a. It has a stochastic time trend, so we can de-trend it the same way we de-trend time series with deterministic time trend. b. First differencing a random walk generates a stationary time series. c. It shows explosive behavior. d. Its highly consistent behavior comes from the fact that it transfers past to future without diminishing its effect.

bd

22. Which one is correct about the basic Dickey-Fuller test? (Two correct answers.) a. The null hypothesis is that the variable is stationary. b. It tests for the existence of unit root c. Rejection of the null hypothesis implies that the variable has explosive behavior. d. It is a one-sided test and the rejection region is in the left-hand side.

bd

14. Which one is the idea of tests of detecting heteroscedasticity such as Bruesch-Pagan test? (One correct answer.) a. Regress squared residuals on explanatory variables and test the significance of each explanatory variable. b. Regress residuals on explanatory variables and test the overall validity (significance) of the entire model. c. Regress squared residuals on explanatory variables and test the overall validity (significance) of the entire model. d. Regress residuals on explanatory variables and test the significance of each explanatory variable.

c

28. Which one is correct about different estimators using panel data? Check all that apply. (Two correct answers.) a. Pooled cross section estimators fully utilize all the information a panel data contains. b. The results of pooled cross-section are always less reliable then the results of fixed-effects models, because we do not use the time-related information. c. Dummy variable estimators cannot be applied in cross-sectional data because of the number of variables cannot be more than the number of observations. d. Dummy variable estimators are technically the same as fixed-effects estimators.

cd

7. Which one is correct about linear probability model? Check all that apply. (Two correct answers.) a. Its interpretation is based on the probability of dependent variables being zero. b. It always predicts a probability that is between zero and one. c. It is estimated using the OLS method. d. Its simplicity is so appealing that many prefer to use it despite its shortcomings.

cd

26. Which one is correct about panel data? Check all that apply. (One correct answer.) a. Observable time-invariant factors cannot be included in panel data analysis, because they do not change over time. b. Every explanatory variable in panel data models necessarily varies over time and among individuals. c. In every panel data, the time span, T, is smaller than the number of individuals in each period, n. d. Unobservable time-invariant factors do not change over time, so we can control for them in panel data.

d

3. Which one is correct about using dummy variables in regressions? Check all that apply. (One correct answer.) a. Adding unrelated dummy variables to the model reduces R-squared, so it is not useful. b. We can use as many dummy variables as we want, because unlike numerical variables, they don't reduce degrees of freedom. c. We cannot use only dummy variables as explanatory variables. There should be at least one numerical variable in the model. d. The number of dummy variables we use in a regression is limited only by the number of observations, i.e., degrees of freedom.

d