gis spatial statistics for public opinion survey response rates · 2015-07-10 · the gis industry...
TRANSCRIPT
GIS Spatial Statistics for Public Opinion Survey Response Rates
Timothy MichalowskiSenior Statistical GIS AnalystAbt SRBI - New York, [email protected]
July 22, 2015
Abt SRBI | pg 2
Introduction§ Timothy Michalowski, Senior Statistical GIS Analyst§ Abt SRBI in New York City
§ “GIS Manager” - Abt SRBI GIS group in the Advanced Methods department, formed in 2010.
§ 15 years GIS experience, focus on GIS for Social Research.
§ Abt SRBI - www.srbi.com
§ One of the nation’s largest and most trusted survey, opinion, and policy research organizations.
§ Abt SRBI offices in NYC, Cambridge, DC area, Chicago, Arizona, North Carolina, & Florida.
§ Expertise in many practice groups, including Health, Transportation, Social Policy, Housing, Energy, Election Polling, etc.
Abt SRBI | pg 3
Introduction
Goals for today
§ Keep everyone awake, it is 8:30am!
§ Why is GIS important to Survey Research?
§ Sample of GIS work at Abt SRBI for Survey Research
§ Geographic Based Statistics for Response Rates
§ Get you out of here for SD Padres 12:40pm game!!!!
Abt SRBI | pg 4
GIS for Survey Research
For 2012 to 2015:
§ “the GIS market in the US to grow at a Compound annual growth rate of 10.96%”
§ “the key growth is due to increased demand from Government sectors.”
The GIS industry grew worldwide 10.3% in 2010 to US $4.4 billion with an and was expected to US $10.6 Billion in 2015.
Global Industry Analysts (GIA), Inc., “Geographic Information Systems (GIS): A Global Outlook,” January 2012.
Abt SRBI | pg 5
GIS for Survey ResearchThe fastest growing segment of GIS from 2004 to 2011 was in GIS data, which grew at a compound annual rate of 15.5%, twice the rate of growth for GIS software and services.
Global sales to governmental agencies remained had a 7.2% compound annual growth rate over these eight years.
“…demand for GIS/geospatial products is driven by an increasing global need for geographically correlated information. As more and more websites, such as Google Earth, and consumer navigation systems, bring awareness of the power of linking business and consumer information with their geography, geo applications will become the norm.”
Bridgett Pelosi, Daratech, Inc, “GIS/Geospatial Market Report," January 2011, available from Directions Magazine, accessed March 28, 2015.
Abt SRBI | pg 6
Survey of GIS Survey Researchers
Abt SRBI GIS completed a Survey of GIS Survey Researchers (March 2015)
§ The target population for the survey was peer GIS professionals at fellow survey research organizations.
§ Developed a short survey instrument to gauge immersion of GIS in the survey research marketplace.
§ The short web survey was distributed to comparable major survey research competitors.
Abt SRBI | pg 7
Survey of GIS Survey Researchers
Avg. Number of GIS Employees
To All Employees
(2%)
(46%)
Abt SRBI | pg 8
Survey of GIS Survey Researchers
Abt SRBI | pg 9
Survey of GIS Survey Researchers
Abt SRBI | pg 10
Survey of GIS Survey Researchers
Abt SRBI | pg 11
Survey of GIS Survey Researchers
Abt SRBI | pg 12
Abt SRBI GIS for Survey Research
§ Surveys always contain a “where”.– Respondent Location (Street, Zip Code, City, State, County,
Census geographies, neighborhoods, customized zones)
– Respondent Other Locations (Work, School, Shopping, Recreation, Transit Stations)
– Location of a Study Entity: (Powerplant, Coastline, Bank, Commercial Location, etc)
– Perceived Locations: (Where is _____?)
§ There exists a geographic element to every survey for potential analysis.
§ Abt SRBI GIS leverages the “where” in surveys
Abt SRBI | pg 13
GIS for Survey Research
§ Proposals
§ Study Areas
§ Simple Maps
§ Demographics
§ Survey Targets
§ Sample Plans
§ Data Collection
§ Field Surveys
§ Geocoding
§ Monitoring
§ Quality Checks
§ Analysis
§ Visualization
§ Web Maps
§ GeoStatistics!
Start Main Study End
Basic uses of GIS
Advanced usesof GIS
Abt SRBI | pg 14
Study Area Boundaries
Abt SRBI | pg 15
Study Area Boundaries
Abt SRBI | pg 16
Study Area Boundaries
Abt SRBI | pg 17
Complex Study Areas
Abt SRBI | pg 18
Complex Study Areas
CustomizedTarget Areas
Abt SRBI | pg 19
Demographics of Study Areas
Abt SRBI | pg 20
Demographics of Study Areas
SurveyTarget
Populations?
Abt SRBI | pg 21
Sample Plans – Target Areas
Oversampling of
TransitPropensity
Areas
Census Tracts
Abt SRBI | pg 22
Sample Plans – Target Areas
Oversampling of
TransitPropensity
Areas
Census Tracts
Abt SRBI | pg 23
Address Based Sample & GISIncrease in cell
phone only households
requires Address Based Sample
(ABS) vs. traditional landline
approach
GIS allows for customized survey target plans w/ABS
Addresses directly in Tornado Paths
Abt SRBI | pg 24
Address Based Sample & GIS
Address Based Sample Stratification Plan by Area in
CT Evacuation Zones & Census Block Groups
Abt SRBI | pg 25
Address Based Sample & GIS
Address Based Sample based on Stratification Plan
Abt SRBI | pg 26
Mapping Survey Results
Abt SRBI | pg 27
Mapping Survey Results
Abt SRBI | pg 28
Mapping Survey Results
Abt SRBI | pg 29
GEOGRAPHIC STATISTICS!
Abt SRBI | pg 30
Why are “GeoStatistics” important for Survey Research?§ Survey plans/sample plans look at a geographic
distribution on a broad level statistically, weight those large areas
§ For general survey administration, assumptions made on a homogeneous spread within the areas
§ Surveys expect the selected study sample to reflect the population distribution and mirror the natural clustering.
§ GeoStatistics helps to either confirm the natural pattern or reveals departures from expected.
Abt SRBI | pg 31
Why are “GeoStatistics” important for Survey Research?
§ Statistically significant geographic patterns of data:
– Examination of survey participation rates to geography
– Location of where surveys are completed, from survey distribution locations, could impact survey participation
– Geographic clustering can be considered in survey design, sample plans, response rates, etc.
– Spatial patterns of survey results help explain findings
– Geographic areas could be oversampled for a study, just as certain demographics are
Abt SRBI | pg 32
What are “GeoStatistics”? Spatial AutoCorrelation
Positive Spatial No spatial Negative spatialautocorrelation autocorrelation autocorrelation
CLUSTERED RANDOM DISPERSED(+) (0) (-)
Statistical tests to measure significance of spatial autocorrelation
Source: Data Urbanism. Dataurbanist.com
Abt SRBI | pg 33
Geostatistics Methods / ToolsZ scores & P values
§ Z score – measureof standard deviation
§ P value – measure of significance
Source: ESRI, resources.arcgis.com
Abt SRBI | pg 34
Geostatistics Methods / ToolsAnselin Global Moran’s I
§ “Moran’s I” is used to measure “spatial autocorrelation”, i.e. “clustering”
§ “Global Moran’s I” Determines whether all featurespart of a statistically significant cluster§ Tests if overall pattern is randomly distributed
(null hypothesis), dispersed, or clustered§ Assigns a “Z score” & “P value” for all data
GLOBAL STATISTIC
A measurement of all data
points
Abt SRBI | pg 35
Geostatistics Methods / ToolsAnselin Local Moran’s I§ “Moran’s I” is used to measure “spatial
autocorrelation”, i.e. “clustering”
§ “Local Moran’s I” Determines whether each feature is part of a statistically significant cluster§ Identifies statistically significant hot spots, cold
spots, and spatial outliers§ Assigns a “Z score” & “P value” for each data point
LOCAL STATISTIC
A measurement of each data
point
Abt SRBI | pg 36
Example 1:GPS Unit Retrieval
§ GPS Units sent to households in Minneapolis for a Travel Survey
§ % of Households returned GPS Units with valid travel data
§ Q: Is there a geographic pattern to GPS travel study compliance?
Source: Abt SRBI
Abt SRBI | pg 37
Example 1:GPS Unit Retrieval
§ Map the household locations with value of compliance vs. non-compliances
§ Use of Interpolation first to visualize spatial patterns –Inverse Distance Weighting (IDW).
§ Clusters do appear of non-compliance but are there statistically valid patterns?
Source: Abt SRBI
Abt SRBI | pg 38
Example 1:GPS Unit Retrieval
§ Z score – measureof standard deviation
§ P value – measure of significance
Source: ESRI, resources.arcgis.com
Abt SRBI | pg 39
Example 1:GPS Unit Retrieval§ Tests INDIVIDUAL data point for clustering, part of a statistically
significant cluster (“hot spot”) for the variable "GPS unit return compliance”
§ Each point receives a Z score and P value
§ RESULTS = Most Z scores < 1.65 and all P values > 0.1NOT SIGNIFICANT
LOCAL Moran’s I
GLOBAL Moran’s I§ Tests OVERALL dataset for clustering
§ Overall Z score and P value
§ RESULTS = Z score = 0.78, P value = 0.43NOT SIGNIFICANT
Abt SRBI | pg 40
Example 1:GPS Unit Retrieval
§ Age / household size are significantly correlated with GPS compliance
§ Household location not significantly correlated with GPS compliance
§ Age and household size are the most significant predictors of GPS compliance, not geographic location
One-way ANOVA – Demographic Data
Source: Abt SRBI
RESULTS:
Abt SRBI | pg 41
Example 1:GPS Unit Retrieval
ESRI UC Map Gallery 2015!
Abt SRBI | pg 42
Example 1:GPS Unit Retrieval
ESRI UC Map Gallery 2015!
Abt SRBI | pg 43
Example 2:Survey Response Rate Coastline
§ Survey of households with fishing licenses in Southern California
§ % response rate for completed surveys
§ Q: Is there a geographic pattern to response rate?
§ Q: Does the response rate vary by distance from coastline?
Source: Abt SRBI
Abt SRBI | pg 44
Example 2:Survey Response Rate From Coastline
§ Interpolation of data points, households with completed surveys (green) vs. non-response (red)
§ Can be used to predict response rate in non sampled areas
§ Reveals a general pattern of clustering of completed surveys near coastline (green)
§ Statistically significant?Source: Abt SRBI
Abt SRBI | pg 45
Example 2:Survey Response Rate From Coastline
§ “Directional Distribution”
§ Mapping “distributional trend”
§ Elliptical polygons centered on the mean centers:Households with completed surveys vs. non-response
Source: Abt SRBISource: ESRI, resources.arcgis.com
Abt SRBI | pg 46
Example 2:Survey Response Rate From Coastline
§ Global Moran’s IZ score = 2.86P value = 0.004
§ Strong significance of spatial autocorrelation “clustering”
§ Correlation between survey response and distance from coastline Source: Abt SRBI
Results:
Abt SRBI | pg 47
Example 3:Foreclosure survey sample selection
§ A survey for homeowners in neighborhoods with highest foreclosure rates. National dataset of foreclosures - Where to conduct interviews?
§ Q: Where are significant neighborhoods (“clusters”) of foreclosures?
§ Anselin Local Moran’s I for all foreclosure locations to find points with P<.05 and highest Z scores selected as central points for sampling neighborhood interviews.
Abt SRBI | pg 48
Example 3:Foreclosure survey sample selection
§ Anselin Global Moran’s I used for all foreclosure locations to verify that the entire dataset has a Z score of clustering and P value of significance.
§ Getis Ord to examine concentrations of hot spots and cold spots.
Abt SRBI | pg 49
Example 3:Foreclosure survey sample selection
Source: Abt SRBI
§ Data points (foreclosure locations) with P<.05 and highest Z scores selected as central points for sampling neighborhood interviews.
Abt SRBI | pg 50
Example 4:Field survey location analysis
§ A survey completed of voting behavior
§ Completed survey data lists addresses / persons who voted (green) or did not vote (red)
§ Corresponding polling places are known (yellow)
§ Q: Is there a significance between polling location andvoting behavior?
Source: Abt SRBI
Abt SRBI | pg 51
Example 4:Field survey location analysis
§ Geographically Weighted Regression (GWR) / Exploratory Regression Analysis
§ Dependent Variable:Voted / Did Not Vote
§ Explanatory Variable:Distance to polling location
§ RESULT: Through geographic based statistical tests, interpretation of results, distance to polling place shows a significant correlation on voting behavior
Abt SRBI | pg 52
Example 4:Field survey location analysis
§ Reverse: Could be used to also measure significance of field survey interviewing locations (customer satisfaction surveys).
§ Do respondents who live farther from survey locations show correlation to non-response?
§ A tool to better establish survey distribution locations for potential respondents
Source: Abt SRBI
Abt SRBI | pg 53
Conclusions
§ Collected survey data has an inherent location (respondent address, survey location, etc.)
§ GIS adds a dimension to existing survey data with advanced analysis via geographic based statistics
§ Geographic statistics can be used to help better plan for and optimize surveys, examine sample address data, regional response rates, clusters of data, and management of field survey locations, etc.
Abt SRBI | pg 54
Questions?
Thank you!
Timothy MichalowskiSenior Statistical GIS Analyst
Abt SRBI275 Seventh Ave, Suite 2700
New York, NY 10001