Presentation is loading. Please wait.

Presentation is loading. Please wait.

Ken Hodges January 2014 IMPROVING THE ACCURACY OF BLOCK GROUP DATA FROM THE AMERICAN COMMUNITY SURVEY.

Similar presentations


Presentation on theme: "Ken Hodges January 2014 IMPROVING THE ACCURACY OF BLOCK GROUP DATA FROM THE AMERICAN COMMUNITY SURVEY."— Presentation transcript:

1 Ken Hodges January 2014 IMPROVING THE ACCURACY OF BLOCK GROUP DATA FROM THE AMERICAN COMMUNITY SURVEY

2 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 2 BACKGROUND ACS data for block groups Concern over accuracy Large margins of error Conspicuous outliers Census Bureau does not release on American FactFinder Only on ACS Summary File ACS sample Smaller than census long form Many estimates based on few responses

3 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 3 BACKGROUND Block Groups by Number of ACS Responses (2006-2010) ACS ResponsesNPercent Missing3800.2 Less than 102,6501.2 10 to 1918,2648.4 20 to 49132,66960.9 50 to 9953,31524.5 100 to 1999,5554.4 200 to 4998960.4 500 or more110.0 Total217,740100.0

4 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 4 BACKGROUND Accuracy much better for aggregations But some d ismiss value of block group data Critical of availability More productive to seek improvements Paper describes: ACS Touch Method Effort to improve accuracy Before use in Nielsen products Also evaluation against 2010 census data

5 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 5 ACS TOUCH METHOD Effort to improve accuracy of ACS block group data Augment with data from adjacent block groups Expectation: adjacent BGs tend to be similar A form of “borrowing strength” Similar to census allocation and substitution Requires mega data processing But conceptually and computationally simple

6 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 6 ACS TOUCH METHOD

7 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 7 ACS TOUCH METHOD

8 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 8 ACS TOUCH METHOD For each ACS table and each block group Three distributions ACS Published: Data as published in ACS Summary File ACS Touch: Published combined with adjacent BGs ACS Weighted: Weighted avg of Published and Touch Weight based on Number of ACS responses If 100+ responses, Published weighted 100 pct Touch weight increased as responses decrease

9 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 9 2006-2010 ACS PERCENT DISTRIBUTION OF HOUSEHOLD TYPE BY HOUSEHOLD SIZE: BG 16 001 0004.00 1 IN IDAHO HH Type and SizeACS PublishedACS TouchACS Weighted Total Households3284,082328 Family 2 persons36.920.123.3 Family 3 persons8.59.89.6 Family 4 persons21.312.514.2 Family 5 persons3.73.13.2 Family 6 persons0.01.91.6 Family 7+ persons0.0 Nonfamily 1 person3.439.732.8 Nonfamily 2 persons4.08.97.9 Nonfamily 3 persons0.02.21.7 Nonfamily 4 persons0.0 Nonfamily 5 persons0.0 Nonfamily 6 persons0.0 Nonfamily 7+ persons22.31.85.7 Unweighted HU19

10 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 10 EVALUATION Key question Does this approach make data more (or less) accurate? Evaluation Looked at “Households by Type and Size” Available for both ACS and census Compared all 3 ACS distributions vs. census 2006-2010 data for all BGs Index of dissimilarity An imperfect but revealing comparison

11 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 11 2006-2010 ACS PERCENT DISTRIBUTION OF HOUSEHOLD TYPE BY HOUSEHOLD SIZE: BG 16 001 0004.00 1 IN IDAHO HH Type and Size2010 CensusACS PublishedACS TouchACS Weighted Total Households3763284,082328 Family 2 persons22.336.920.123.3 Family 3 persons17.68.59.89.6 Family 4 persons7.721.312.514.2 Family 5 persons5.13.73.13.2 Family 6 persons1.90.01.91.6 Family 7+ persons0.50.0 Nonfamily 1 person32.23.439.732.8 Nonfamily 2 persons11.44.08.97.9 Nonfamily 3 persons1.10.02.21.7 Nonfamily 4 persons0.30.0 Nonfamily 5 persons0.0 Nonfamily 6 persons0.0 Nonfamily 7+ persons0.022.31.85.7 Index of Dissimilarity 50.415.214.4

12 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 12 EVALUATION OK... the big outlier is improved What about across all block groups? Mean Index of Dissimilarity for All Block Groups: Alternative ACS Distributions versus 2010 Census Does this differ by N of ACS responses? ACS ComparisonMean IOD ACS Published vs. Census19.4 ACS Touch vs. Census12.2 ACS Weighted vs. Census12.2

13 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 13 MEAN INDEX OF DISSIMILARITY FOR BLOCK GROUP ACS DISTRIBUTIONS OF HOUSEHOLD TYPE BY SIZE BY NUMBER OF ACS RESPONSES ACS ResponsesNPublishedTouchWeighted All Block Groups*216,59819.412.2 Fewer than 101,90060.332.432.1 10 to 1918,25529.315.014.8 20 to 2950,29123.213.113.0 30 to 3948,37919.412.112.0 40 to 4933,99616.911.511.4 50 to 5922,11415.311.111.0 60 to 6913,88414.210.710.9 70 to 798,51213.410.411.1 80 to 895,31412.710.211.3 90 to 993,49112.210.011.7 100 or more10,46210.89.210.8 * Block groups with 2010 households greater than 0.

14 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 14 EVALUATION “Touch” better than “Published” even where larger N of ACS responses Better to use ACS Touch for all BGs? Another view How often do Touch and Weighted improve over Published?

15 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 15 PERCENT OF BLOCK GROUPS WHERE ACS TOUCH AND WEIGHTED HAD LOWER MEAN INDEX OF DISSIMILARITY THAN ACS PUBLISHED: HOUSEHOLD TYPE AND SIZE BY NUMBER OF ACS RESPONSES ACS ResponsesNTouchWeighted All Block Groups*216,59882.588.4 Fewer than 101,90091.292.1 10 to 1918,25592.995.3 20 to 2950,29189.494.0 30 to 3948,37985.292.8 40 to 4933,99680.591.8 50 to 5922,11477.391.3 60 to 6913,88474.291.6 70 to 798,51272.092.4 80 to 895,31470.493.5 90 to 993,49168.293.0 100 or more10,46265.40.0 * Block groups with 2010 households greater than 0.

16 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 16 EVALUATION OK... improvement across BGs But is there a negative impact at aggregate levels? Mean Index of Dissimilarity for ACS Distributions: Block Groups Summed to County and National Levels Conclusion: No major negative impact on aggregate areas ACS ComparisonCountyNational ACS Published vs. Census4.82.0 ACS Touch vs. Census4.61.8 ACS Weighted vs. Census4.72.1

17 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 17 CONCLUSIONS Comprehensive test not feasible ACS is only source for most data it provides ACS period estimates not comparable with census Limited evaluation suggests potential ACS outliers improved Mean IODs are reduced Improved accuracy in most BGs Accuracy of aggregations preserved

18 Copyright ©2013 The Nielsen Company. Confidential and proprietary. 18 CONCLUSIONS Could debate merits of Touch vs. Weighted But the potential for improvement is clear And the application could be refined More selective designation of “touch” BGs Those similar on decennial characteristics The choice is ours... Dismiss BG data as error-prone Or seek to reduce error, and maximize value

19 Thank You ! Ken.hodges@nielsen.com


Download ppt "Ken Hodges January 2014 IMPROVING THE ACCURACY OF BLOCK GROUP DATA FROM THE AMERICAN COMMUNITY SURVEY."

Similar presentations


Ads by Google