Update: Results of Simulation Study for the EIA-914

1
Update Results of Simulation Study for the
EIA-914

• Preston McDowney
• Statistics and Methods Group
• U.S. Department of Energy
• preston.mcdowney_at_eia.doe.gov

2
2004 Spring Meetings Recap

• Inderjit Kundra
• EIA was in the process of developing a new
  survey, the EIA-914, to collect monthly
  production of natural gas in the United States
• EIA was considering a Probability Proportion to
  Size (PPS) sample.
• The goals were to estimate total US production
  with a coefficient of variation (CV) of 1 and
  regional production with a CV of 5.

3
Regions
2004 Spring Meetings Recap

• Texas (TX)
• Federal Gulf (FG)
• Louisiana (LA)
• New Mexico (NM)
• Oklahoma (OK)
• Wyoming (WY)
• Others (excluding Alaska)

4
How
2004 Spring Meetings Recap (cont.)

• Sample size was determined by using Presumed
  Optimum Allocation formulae

Where CSS denotes calculated Sample
Size h denotes stratum in a region N(h) denotes
number of operators in a stratum S(h) denotes
standard Deviation in stratum h CV denotes
coefficient of variation y(hi) denotes the
production for operator i in stratum h
5
Estimated Sample Sizes
1 An operator within a region having a measure of
size greater than or equal to total production
divided by 2n was selected with certainty.
6
Four Estimation Procedures
2004 Fall Meeting Recap

• Probability Proportion to Size (PPS) Estimator
• Weighted Least Squares (WLS) Estimator
• Ordinary Least Squares (OLS) Estimator
• Difference Estimator

7
Variations of the WLS and OLS

• Two Methods
• Using the non-certainty group only in the
  estimation procedure
• Using both the certainty and non-certainty groups
  in the estimation procedure
• Three Variations
• Using all of the operators in the method
• Removing outliers
• Removing outliers and overly influential
  observations

8
Simulation Results,Percent Error Estimates for
Total Natural Gas Production
2004 Fall Meeting Recap (cont.)
9
Fall 2004 Conclusions

• The PPS sampling did not provide the desired
  accuracy
• ASA Committee Recommended Using the Hajek
  Estimator
• Decided to use a cut off sample with 90 percent
  coverage at the national level
• John Wood Presented the need for a Bias
  Adjustment

10
Spring 2005 Meeting Recap

• John Wood Presented Preliminary January and
  February 2005 Data
• Demonstrated the Estimation Procedure and
  Adjustment For Negative Bias
• Reported 99.4 Response Rate (Volume Weighted)

11
Stage 2- Major Differences

• New Frame
• Hajek Estimator
• Stratified Sample
• Focus on 2 regions TX and FG

12
Frame Details

• Stage 1
• Annual Data
• Sources
• EIA-23 Survey Data
• State Data
• Multiple Third Parties

• Stage 2
• Monthly Data
• Sources
• Single Third Party

13
Stage 1
Regional Detail
Stage 2

• Integrated
• Texas (TX)
• Federal Gulf (FG)
• Louisiana (LA)
• New Mexico (NM)
• Oklahoma (OK)
• Wyoming (WY)
• Others (excl. Alaska)

• Individual
• Texas (TX)
• Federal Gulf (FG)

14
Sample Sizes

• Stage 1

Stage 2
15
Sampling Methodologies and Estimators

• Probability Proportion to Size (PPS)
• PPS Estimator
• Weighted Least Squares (WLS)
• Ordinary Least Squares (OLS)
• Difference Estimator
• Hajek Estimator
• Random Stratified Sample
• Stratified Sample Estimator

Model Based Methods
16
Hajek Estimator

• YEstimate
• wi Weight
• yi Observation
• nnon-certainty sample size

17
Stratified Sample

• Random Stratified Sample with Replacement
• 6 non-certainty strata in Texas
• 2 non-certainty strata in the Federal Gulf

18
Review of Summary Results

• Compare the Percent Errors of the Estimated
  Regional Annual Totals (summation of monthly
  estimates)

19
Texas Stage 2 Annual Percent Error of
Estimated Total
20
Federal Gulf Stage 2 AnnualPercent Error of
Estimated Total
21
Conclusions

• In General
• Probability Methods Had Largest Variation And In
  Many Cases Negative Bias
• Difference Estimator and Model Based Estimators
  with Certainty and Non-Certainty Operators Had
  the Smallest Variation and Bias
• Outlier and Influential Observations Detection
  Made Little Difference

22
(No Transcript)
23
Texas New Simulation Annual ResultsPercent Error
of Estimated Total (outliers and influential
variables removed)