Presentation is loading. Please wait.

Presentation is loading. Please wait.

Incorporating Nonresponse in a Markov Latent Class Measurement Error Model of Consumer Expenditure Brian Meekins, Clyde Tucker Bureau of Labor Statistics.

Similar presentations


Presentation on theme: "Incorporating Nonresponse in a Markov Latent Class Measurement Error Model of Consumer Expenditure Brian Meekins, Clyde Tucker Bureau of Labor Statistics."— Presentation transcript:

1 Incorporating Nonresponse in a Markov Latent Class Measurement Error Model of Consumer Expenditure Brian Meekins, Clyde Tucker Bureau of Labor Statistics Paul Biemer RTI International Any opinions expressed in this paper are those of the authors and do not constitute policy of the Bureau of Labor Statistics

2 Uses repeated measurements from panel survey data to estimate classification error Used successfully in evaluation of labor force data (e.g., Biemer & Bushery, 2001 Tucker, Biemer, Meekins 2012) Does not require external validation data; estimates of error directly from panel data Applied to reports of expenditures in the CEIS Markov Latent Class Analysis

3 U.S. Consumer Expenditure Interview Survey (CEIS) ~ 6,000 CU’s/year CU’s interviewed every 3 months about prior 3 months expenditures 4 consecutive interviews on each CU New data: N=23,719 From 2005.2 to 2009.2 Unweighted analysis

4 ElectricityMen’s Clothing OnlyMinor Vehicle Repairs CableWomen’s Clothing OnlyVehicle Oil TrashMen’s AccessoriesMajor Vehicle Repairs GasWomen’s AccessoriesVehicle Other TVMen’s ShoesDental SportsWomen’s ShoesEyes FurnitureKid’s ApparelDrugs* Other HouseholdKid’s Clothing OnlyPets Kitchen AccessoriesKid’s Accessories Men’s ApparelKid’s Shoes Commodity Categories 4

5 1 st Order Markov Model 5 ZYX W Where

6 Definition of Indicator Variables Define for Interview 1, 1, if reported as a purchaser for the quarter 2, if reported as non-purchaser with similar definition for X, Y, Z for 2nd, 3rd, and 4th interviews W=

7 2 nd Order Markov Model 7 ZYX W

8 Mover-Stayer 8 ZYX W M 1, P(W=1) = P(X =1) = P(Y =1) = P(Z = 1) = 1 M = 2, P(W=1) = P( X =1) = P(Y =1) = P(Z = 1) = 0 3, P(W),P(X), P(Y), and P(Z) are unconstrained.

9 Measurement Error Model 9 ADCB ZYX W = = = =

10 Markov Latent Class Model Notation True expenditure status is a latent variable Latent Var. Indicator Var. Interview 1W A Interview 2X B Interview 3Y C Interview 4Z D

11 Definition of Latent Variables Where, 1, if one or more purchases of an item during the W=quarter (“purchaser”) 2, if no purchase (“non-purchaser”) with similar definition for X, Y, Z for 2nd, 3rd, and 4th interview

12 Markov or Mover-Stayer model assumptions Equal measurement error across all interviews No False Positives Model Assumptions

13 Model Selection Limitations on Lem forced estimation of separate pieces of the full model: ME & NR Multiple iterations to avoid local maxima Best refusal, noncontact, and measurement error variables were selected based on fit in these component models Using best variables constructed combined model

14 Estimates ME Model No covariates P(A=1|W=1) = P(B=1|X=1) = P(C=1|Y=1) = P(D=1|Z=1) = 0.763 14

15 Measurement Error Model 15 ADCB ZYX W = = = = Measurement Error Indicators

16 Measurement Error Model 16 ADCB ZYX W Measurement Error Indicators

17 ME Variables 1. Missing income data (CU) 2. Type and frequency of records used (interiew) 3. Length of interview (interview) 4. Ratio of expend. in last month to entire quarter (interview) 5. Combination of types of record and int length (interview) 6. Number of expenditure questions imputed/allocated (interview) 7. Completion mode (interview) 8. Family size (CU)

18 ME Model w/Covariates 18 ME Model w/ Covariates P(A=1|W=1)0.781 P(B=1|X=1)0.733 P(C=1|Y=1)0.696 P(D=1|Z=1)0.687

19 ME Model: Covariates 19 Income MissingP(A=1|W=1) Missing0.779 Not Missing0.796 ModeP(A=1|W=1) Telephone0.766 In person0.814 Record use/Int timeP(A=1|W=1) Short int/few records0.664 Medium int/med records0.779 Long int/lots of records0.982

20 Nonresponse Error Model 20 EHGF Noncontact Covariates 1 Reports All Qtr 2 Reports Some 3 Reports None M Refusal Covariates

21 Nonresponse Error Model 21 EHGF Noncontact Covariates 1 Latent.. likelihood c of Interview M Refusal Covariates

22 Noncontact Variables 1. Number of noncontact problems reported in CHI (interview) 2. Age (CU) 3. Owner/Renter (CU) 4. Urbanicity (CU)

23 Noncontact Estimates 23 WaveProb of Responding E1.000 F0.877 G0.823 H0.798

24 Noncontact Estimates 24 Noncontact Problems P(F=1) % Samp No mentions0.864 39.8 1 or 2 mentions0.945 33.8 Three or more0.714 26.4 P(F=1) % Samp Rents0.812 32.1 Owns0.905 67.9

25 Noncontact Estimates 25 AgeP(F=1) % Samp <300.779 13.8 30-490.870 39.7 50 +0.912 47.1 UrbanicityP(F=1) % Samp Urban0.874 94.2 Rural0.905 5.8

26 Refusal Variables 1. Factor1 variables – Reluctance privacy concerns (CU) 2. Factor2 variables – Reluctance time concerns (CU) 3. Any reluctance mentioned (CU) 4. Conversion Refusal – only wave 1 available? (CU) 5. Region (CU) 26

27 Refusal Estimates 27 Refusal Reason Factor 2P(M=3) Worst % Samp Never mentioned0.086 42.8 Avg 1 per interview0.034 34.1 More0.152 23.1 Refusal Reason Factor 1P(M=3) Worst % Samp Mentioned0.144 26.1 Not mentioned0.072 73.9

28 Refusal Estimates 28 RegionP(M=3) Worst % Samp Northeast0.077 18.9 Midwest0.078 23.4 South0.075 35.3 West0.107 22.4 Any Conversion RefusalP(M=3) Worst % Samp Yes0.136 12.2 No0.076 87.8

29 29 EHGF Noncontact Covariates Refusal Covariates ADCB ZYX W = = = = Measurement Error Indicators 1 Reports All Qtr 2 Reports Some 3 Reports None M Combined ModelCombined Model

30 Estimates: Full Model Measurement Error (Item Nonresponse) P(A=1|W=1)0.781 P(B=1|X=1)0.758 P(C=1|Y=1)0.739 P(D=1|Z=1)0.745 Unit Nonresponse P(F=1|X=1)0.957 P(G=1|Y=1)0.953 P(H=1|Z=1)0.957

31 Missing Reports Of the estimated 1,939 missed reports  1,721 were a result of item nonresponse  218 were a result of unit nonresponse Reminder: nonresponse conditioned on first wave response 31

32 Discussion Can be conducted for each commodity category Models need further specification & adjustments, thorough evaluation of covariates Estimates can be used to inform:  Allocation of resources  Adjustment: relationship of NR and ME  Fatigue, conditioning, etc. 32

33 Contact Information Brian Meekins Office of Survey Methods Research U.S. Bureau of Labor Statistics 202-691-7594 meekins.brian@bls.gov

34 Objective Diagnositics Fit Statistics  L-square Dissimilarity Index BIC

35 Subjective Diagnostics True purchase pattern given estimated purchase pattern Mover-stayer classification by purchase pattern Accuracy rates by subgroup  Accuracy is the percent of true purchasers that reported purchasing that commodity


Download ppt "Incorporating Nonresponse in a Markov Latent Class Measurement Error Model of Consumer Expenditure Brian Meekins, Clyde Tucker Bureau of Labor Statistics."

Similar presentations


Ads by Google