Presentation is loading. Please wait.

Presentation is loading. Please wait.

Measurement Error in Survey Data: Revisiting the Study of Income and Consumption Dynamics Nayoung Lee University of Southern California 2008 North American.

Similar presentations


Presentation on theme: "Measurement Error in Survey Data: Revisiting the Study of Income and Consumption Dynamics Nayoung Lee University of Southern California 2008 North American."— Presentation transcript:

1 Measurement Error in Survey Data: Revisiting the Study of Income and Consumption Dynamics Nayoung Lee University of Southern California 2008 North American Summer Meeting of the Econometric Society, Pittsburgh

2 Economic Mobility “It is mobility which provides the sticks for those who do not wish to move down the distribution and the carrots for those who wish to move up. ” - Hart (1981)

3 Measurement Error in Survey Data No way! My answer will be compared to my tax file! How much did I earn and spend last year? I have no idea! I don’t wanna be treated like a looser. I am going to answer it higher! Toolhire Toolhire 2007

4 Outline Objectives Data Empirical Specification Estimation Strategies Results Tests Error Decomposition Conclusion

5 Income and Consumption Dynamics shows the degree how much income (or consumption) at present is determined by income (or consumption) in the past. examines whether people can demonstrate the ability to change their income and move in and out of poverty.

6 Potential Source of Measurement Error Substantial recall error or recording error Intentional underreporting (or overreporting) - income tax, self-respect and etc. Many studies confirm the existence of measurement error in survey data. Mellow and Sider (1983), Bound and Krueger (1991), Codar (1991), Bound, Brown, Duncan and Rodgers (1994), Pischke (1995), Bound, Brown, and Mathiowetz( 2001), Gibson (2002), Attanasio, Battistin and Ichimura (2004), Gottschalk and Huynh (2005), and etc.

7 Objectives The purpose of this study: Identifying measurement error in reported income and consumption Examining the direction and magnitudes of biases generated by potential measurement error in the study of income and consumption dynamics

8 DATA KLIPS (Korean Labor and Income Panel Study) 1998-2006 (2002 to 2005) Yearly per capita household income variable is constructed by four income sources: income from labor, financial assets, real estates, and other income. Transfer income is excluded. It is altered in logarithmic forms after adding 1 (i.e. log(Income+1)). Yearly per capita household consumption variable (directly asked) is altered in logarithmic forms.

9 Empirical Specification Basic Model (1) (2) Measurement Error (3) (4) Model with Measurement Error (5) (6)

10 Empirical Specification Measurement error is decomposed into time-invariant and time-varying components Model after the First Differencing (7) (8)

11 Estimation Strategies (1) ∄ Time-varying Measurement error (2) Time-varying Measurement error Internal (including or ) Internal (excluding or ) or Valid Instrumental Variables for or Identifying Time-varying Measurement Error (Holtz-Eakin, Newey and Rosen (1988)) & External IVs & External IVs ARMA(1,1)

12 External Instrument (S t-2 and S t-3 ) Head of households’ satisfaction regarding their household income Income or Consumption Categories Very Satisfied Very Dissatisfied

13 Results - Income Dynamics Obs: 11,438 Models Level OLS First Differenced Internal and External Instrumental Variables (1) (2)(3) IncludedExcluded Coef. γ 0.524** (0.008) 0.142** (0.026) 0.539 ** (0.153) 0.499* (0.230) ARMA(1,1) ** significant at 1%, * significant at 5% Robust standard error in parenthesis Time-Varying Measurement Error Not Addressed

14 Results - Consumption Dynamics Obs: 11,832 Models Level OLS First Differenced Internal and External Instrumental Variables (1) (2)(3) IncludedExcluded Coef. γ 0.619** (0.007) 0.202** (0.020) 0.406 ** (0.057) 0.403* (0.057) ARMA(1,1) ** significant at 1%, * significant at 5% Robust standard error in parenthesis Time-Varying Measurement Error Not Addressed

15 Hausman Tests Test Null Comparison Hausman Tests Hypothesis IVs Income Consumption γAγA γBγB γ A - γBγB γBγB I AR(1) Specification(3)(2) -0.041 -0.003 (.171) (.003) II No Time-varying(2)(1) 0.397 0.205 Measurement Error (.151) (.053) Standard error in parenthesis

16 Error Decomposition Type of Error ParameterEstimate Income Consumption Equation σεσε 1.067.212 Error (.203)(.031) Measurement σvσv.971.186 Error (.199) (.027) Bootstrap Standard Error in Parenthesis Standard Deviation of the Errors (Equation Error ε and Measurement Error v )

17 Conclusion Substantial time-varying measurement error is in reported income and consumption offsets the combined bias of unobserved heterogeneity and time-invariant measurement error must be considered to avoid counterfactual policies. Without considering time-varying measurement error, income and consumption are much less persistent.

18 Thank you for your attention! I am grateful to John Strauss, John Ham, Jinyong Hahn, Hyungsik Roger Moon, Geert Ridder and Jefferey Nugent for their excellent comments. Corresponding Email Address: nayoungl@usc.edu

19 Control Variables (treated as exogenous ) Household size % of elderly people Educational level of head of household Age of head of household and its square Locality indicator (whether respondent resides in Seoul) Non-spouse indicator (whether HH contains a wife or husband)


Download ppt "Measurement Error in Survey Data: Revisiting the Study of Income and Consumption Dynamics Nayoung Lee University of Southern California 2008 North American."

Similar presentations


Ads by Google