Title: Nonresponse Bias in the 2003 Survey of Small Business Finances
1Nonresponse Bias in the 2003 Survey of Small
Business Finances
- Y. Michael Yang, NORC at the University of
Chicago - Traci Mach, Federal Reserve Board
- Lieu Hazelwood, Federal Reserve Board
2Disclaimer
- The views expressed herein are those of the
authors. They do not necessarily reflect the
opinions of the Federal Reserve Board or its
staff.
3Overview
- Bias due to Nonresponse
- Data Source 2003 Survey of Small Business
Finances - Analytical Methods for Assessing Bias
- Results
- Discussions
4Bias Due to Nonresponse (1)
- Deterministic model
- Probabilistic model
5Bias Due to Nonresponse (2)
- Implications
- Nonresponse bias is variable-specific
- Nonresponse does not necessarily lead to bias
- Nonresponse bias increases as response rate
decreases - Nonresponse bias depends on the response
mechanism - Key is to examine the correlation between
response propensity and the survey variable
62003 SSBF (1)
- Target population
- U.S. for-profit small businesses
- Frame
- Duns Market Identifier (DMI)
- Sample design
- Stratified systematic sampling
- Screener nonresponse subsampling
- Main interview nonresponse subsampling
- Data collection
- Telephone Interview
72003 SSBF (2)
82003 SSBF (3)
- Scope of analysis
- Eligible screener respondents (n6,520)
- Focus on bias due to main interview nonresponse
- Adjustments for screener nonresponse and
subsampling have been incorporated in the base
weight
9Analytical Methods
- Comparing response rates across subgroups
- Comparing early responders and late responders on
survey variables - Comparing first phase responders and follow-up
responders on survey variables - Comparing unadjusted and adjusted estimates
- Estimate nonresponse bias analytically
10Main Interview Outcome Rates
11Outcome Rates by Business Size Unweighted
12Outcome Rates by Business Size Weighted
13Outcome Rates by Urbanicity Unweighted
14Outcome Rates by Urbanicity Weighted
15Early (lt10 calls) Vs. Late (gt10 calls)
Completes Unweighted (1)
16Early (lt10 calls) Vs. Late (gt10 calls)
Completes Unweighted (2)
17Early (lt10 calls) Vs. Late (gt10 calls)
Completes Unweighted (3)
18Early (lt10 calls) Vs. Late (gt10 calls)
Completes Weighted (1)
19Early (lt10 calls) Vs. Late (gt10 calls)
Completes Weighted (2)
20Early (lt10 calls) Vs. Late (gt10 calls)
Completes Weighted (3)
21Early (lt19 calls) Vs. Late (gt19 calls)
Completes Unweighted (1)
22Early (lt19 calls) Vs. Late (gt19 calls)
Completes Unweighted (2)
23Early (lt19 calls) Vs. Late (gt19 calls)
Completes Unweighted (3)
24Early (lt19 calls) Vs. Late (gt19 calls)
Completes Weighted (1)
25Early (lt19 calls) Vs. Late (gt19 calls)
Completes Weighted (2)
26Early (lt19 calls) Vs. Late (gt19 calls)
Completes Weighted (3)
27Main Sample Vs. Follow-up Subsample Completes
Unweighted (1)
28Main Sample Vs. Follow-up Subsample Completes
Unweighted (2)
29Main Sample Vs. Follow-up Subsample Completes
Unweighted (3)
30Main Sample Vs. Follow-up Subsample Completes
Weighted (1)
31Main Sample Vs. Follow-up Subsample Completes
Weighted (2)
32Main Sample Vs. Follow-up Subsample Completes
Weighted (3)
33Indirect Estimate of Bias
- (unadjusted estimate)-(adjusted estimate)
- Measures the amount of nonresponse adjustment
- Measures nonresponse bias if the adjusted
estimate is unbiased
34Indirect Estimates of Bias
Characteristics Estimated Bias Characteristics Estimated Bias
Native American 0.14 Equipment loans -1.27
White 0.07 Capital leases -0.18
White non-Hispanic 0.24 Other loans -0.38
Asian -0.13 Line of credit -0.65
Asian or Pacific Islander -0.1 Transaction service -0.57
Black 0.04 Cash management service -0.51
Pacific Islander 0.03 Credit related service -0.53
Hispanic -0.12 Trust service -0.71
Owner managed 0.77 Brokerage service -0.16
Average owners education 0.02 Card processing service -0.33
Firm age 0.07 Total sales -158570.05
Minority or Hispanic -0.1 Profit -20781.18
Total employment -1.01 Total Assets -77824.28
Mortgage loans 0.07 DB credit percentile -0.96
Motor vehicle loans -1.02 Total employees -1.31
35Direct Estimate of Bias
- Estimate response propensity
- Compute the covariance between response
propensity and survey variable - Estimate bias analytically
36Direct Estimates of Bias (1)
Survey Variable Covariance Estimated Bias Weighted Estimate Bias Ratio
Native American 0.01 0.02 1.41 1.4
White 0.08 0.16 90.82 0.2
White non-Hispanic 0.15 0.31 86.43 0.4
Asian -0.05 -0.11 4.17 -2.6
Asian or Pacific Islander -0.06 -0.11 4.29 -2.6
Black -0.04 -0.07 3.63 -1.9
Pacific Islander 0 0 0.12 0.0
Hispanic -0.09 -0.18 4.28 -4.2
Owner managed 0.15 0.29 94.16 0.3
Average owners education 0.01 0.01 4.61 0.2
Firm age -0.03 -0.05 14.3 -0.3
Minority or Hispanic -0.17 -0.35 13.31 -2.6
Total employment -1.91 -3.85 8.73 -44.1
Mortgage loans -0.11 -0.22 13 -1.7
Motor vehicle loans -0.28 -0.57 25.33 -2.3
37Direct Estimates of Bias Bias (2)
Survey Variable Covariance Estimated Bias Weighted Estimate Bias Ratio
Equipment loans -0.27 -0.54 10.59 -5.1
Capital leases -0.21 -0.42 8.9 -4.7
Other loans -0.21 -0.41 10.08 -4.1
Line of credit -0.53 -1.08 32.45 -3.3
Transaction service -0.4 -0.8 38.96 -2.1
Cash management service -0.55 -1.11 6.43 -17.3
Credit related service -0.25 -0.5 4.82 -10.4
Trust service -0.45 -0.92 17.62 -5.2
Brokerage service -0.05 -0.11 5.45 -2.0
Card processing service -0.16 -0.31 36.67 -0.8
Total sales -331788.2 -669119.94 1,177,572 -56.8
Profit -31853.62 -64239.45 175,965 -36.5
Total Assets -156104.5 -314817.2 535,133 -58.8
38Direct vs. Indirect Bias Estimates
39Discussion
- There is significant correlation between
nonresponse and key survey variables - Larger and urban businesses have lower response
rate - Larger and urban businesses use more services,
have more sales, assets, and profit - Unadjusted estimates of the mean/total are biased
downward for most variables - Response propensity-based weighting appears to be
effective, but only a portion of the bias is
corrected - Weighting method may be improved in future
surveys
40Contact Information