An Application of Spatial Techniques to Audience Measurement Data
Transcription
An Application of Spatial Techniques to Audience Measurement Data
The Power of the Latitude and Longitude: An Application of Spatial Techniques to Audience Measurement Data SORA Business Analytics Seminar May 2013 Derrick Gray, P.Stat. Ricardo Gomez-Insausti, Ph.D. BBM Canada ©2013 BBM Canada Agenda 1. 2. 3. 4. 5. 2 BBM Canada: Background Electronic Audience Measurement Data Types Basic Overview of Spatial Statistics Data-ViU: An Application of Spatial Techniques ©2013 BBM Canada SORA Business Analytics Seminar BBM CANADA: BACKGROUND 3 ©2013 BBM Canada Who is BBM Canada? • BBM Canada, Sondage BBM in Quebec, is a not for profit, member-owned tripartite industry organization, which has been operating since 1944 • BBM Canada provides broadcast measurement and consumer behaviour data, as well as industryleading intelligence to broadcasters, advertisers and agencies through the use of sophisticated data tracking and measurement technologies and services 4 ©2013 BBM Canada National Top 10 TV Programs: Week of May 6-12, 2013 Rank Program Broadcast Outlet Weekday Start End Total 2+ AMA (000) 1 BIG BANG THEORY CTV Total …T… 20:00 20:31 3,117 2 NHL PLAYOFFS ROUND 1 CBC Total MTWTF.S 18:59 21:57 2,935 3 TWO AND A HALF MEN CTV Total …T… 20:31 21:00 2,216 4 NCIS Global Total .T….. 20:00 21:00 2,128 5 SURVIVOR: CARAMOAN Global Total ..W…S 20:00 21:00 1,785 6 GREY’S ANATOMY CTV Total …T… 21:00 22:00 1,735 7 NCIS: LOS ANGELES Global Total .T….. 21:00 22:00 1,616 8 SURVIVOR: CARAM.RERUN Global Total ……S 22:00 23:00 1,602 9 C.S.I. CTV Total ..W…. 22:00 23:00 1,548 10 HAWAII FIVE-O Global Total M…… 22:00 23:00 1,519 BBM Canada: Total Canada, P2+, May 6-12, 2-13 5 ©2013 BBM Canada Toronto Radio Market Rankings: Nov 26, 2012 to Feb 24, 2013 November 26, 2012 – Feb 24, 2013 Average Daily Universe: 5,755,000 6 Rank Call Letters Station 1 CHFIFM CHFI 16.9 5,320.6 1,108.3 2 CHUMFM CHUM FM 10.5 4,790.4 797.7 3 CBLAFM CBC Radio One 10.0 1,808.9 436.7 4 CFTR CFTR 6.5 3,160.5 651.6 5 CKFMFM 99-9 Virgin Radio 6.3 4,611.6 748.7 6 CFRB CFRB 5.8 1,150.6 283.7 6 CILQFM Q107 5.8 2,918.8 459.3 8 CHBMFM boom 97.3 4.8 4,344.3 500.2 9 CFMZF+ The New Classical 96.3 FM 4.7 2,132.4 269.3 10 CKISFM KiSS 92.5 3.7 3,529.1 512.9 ©2013 BBM Canada Share (%) Cume (000) Daily Cume (000) BBM Corporate Structure BBM CANADA 100% BBM ANALYTICS INC 7 ©2013 BBM Canada THE NIELSEN COMPANY 100% BBM Voting Shares BBM ELECTRONIC INC 100% Non-Voting Shares BBM Canada Research Department VP, Research Research & Development Methodology Qualitative Research Project Management Research on Research 8 ©2013 BBM Canada Statistical Research / Research Operations Sample Design & Control Survey Calibration Geographical Design Statistical Analysis Spatial Analysis Call Centre Market Research Division BBM Analytics SORA Business Analytics Seminar ELECTRONIC AUDIENCE MEASUREMENT 9 ©2013 BBM Canada Overview • The Portable People Meter is a cell-phone-sized device with built-in motion detection • It automatically captures inaudible codes embedded in the audio of the television and radio signals • Codes are recorded and time stamped by the meter every 15 seconds • Station with the most codes within a minute is credited with that tuning and each minute is allocated to either television or radio • Time/date stamps allow BBM to differentiate between live and playback viewing 10 ©2013 BBM Canada How the PPM works • Codes are embedded into the audio at the source using encoders provided by BBM: approximately 500 TV stations and 230 radio stations are currently encoded! • All permanent residents within the selected household are assigned a PPM • Bluetooth devices emit a beacon signal which determines whether viewing occurs in or out of the home. If the PPM “hears” this beacon, viewing is attributed to in-home • Approximately 13% of TV viewing and 50% of radio tuning happens outside the home* • At the end of the day, the PPM transmits data to BBM via cellular technology *Source: PPM, Aug 30/10 – Aug 28/11, Total Canada, 2+, AMA(000) 11 ©2013 BBM Canada The PPM System in Action 12 ©2013 BBM Canada PPM Market Samples MONTREAL TORONTO VANCOUVER CALGARY EDMONTON 900 Homes Total 800 Homes Total 672 Homes Total 529 Homes Total 565 Homes Total 800 Radio 600 Radio 450 Radio 400 Radio 400 Radio An Additional 1,004 Homes Round out the Television Panels (Ont Bal., Quebec Bal. Fr., Que Anglo, Atlantic/Prairie/Western Bal.) 13 ©2013 BBM Canada National panel of 4,470 households or approximately 11,000+ people Toronto/Hamilton EM 14 ©2013 BBM Canada Vancouver/Victoria EM 15 ©2013 BBM Canada PPM Methodology – 4 Steps Step Step1: 1:Panel Panel Development Development Step 2: Panel Step 2: Panel Management Management Step 3: Panel Step Assessment 3: Panel Assessment Step 4: 4: Delivery Step Deliv of product product Panel Design Establishment Survey* Recruitment* Install Run-in Period Into Production Editing Weighting* In-Tab Program Schedules Data Delivery *further details on these steps follow 16 ©2013 BBM Canada BBM Canada Establishment Survey • The Establishment Survey is a large scale monthly telephone survey • All telephone numbers are generated using Random B list assisted Random Digit Dialing (RDD) • RDD allows the capturing of both Directory Listed Landline Homes (~85% of Landline Homes) and capturing of Directory Not Listed Landline Homes (~15% of Landline Homes) • Sampling frame is further enhanced to include Mobile only homes (12.8% of Canadian Households and 9.9% of Canadian Individuals*) • Overall frame coverage is 98.9% of Canadian Households and 99.3% of Canadian Individuals* *Statistics Canada, RTSS (2010) 17 ©2013 BBM Canada BBM Canada Establishment Survey continued • Establishment Survey • Approximately 50,000 homes are called each month • The households which complete the survey form the Random probability sample frame of households for later recruitment • The Sample frame is designed to include private households with a landline or mobile-only and individuals who have television and radio access • Questions are limited to those which are required for weighting or panel balancing • The Establishment Survey is a Household survey meaning all questions must be for household variables Information obtained includes limited personal information (official language, age & gender, hours worked) and household information such as radio/TV equipment, reception type, broadcast provider, internet access 18 ©2013 BBM Canada Panel Control / Recruitment • Recruitment to the panel is stratified to ensure the panel is as representative of possible to the Canadian population • Candidates are randomly selected from the Establishment Survey (ES) sample frame based on a set of control variables / strata • These variables are known key contributors to influence different viewing patterns • PPM Panels are multi-media panels (Radio and TV) • Control variables must meet the needs of both media Radio and TV Variables •Geography •Household Size •Presence of demos •Household Moved 19 ©2013 BBM Canada Radio only Variables •Working Status of Secondary Head of Household TV only Variables •Reception Type Panel Calibration • Since not every member of the panel will be intab in a given day, the data is calibrated (poststratified) to correct for any imbalances • Panel Calibration is done using Raking Ratio Estimators using an algorithm called Iterative Proportionate Fit (Deming and Stephan (1940)) • Can also be viewed as a special case of calibration using the Exponential Method. In this case the Newton-Raphson method is needed to solve. Deming, J.-C. and Stephan, C.-E. (1940). On least squares adjustment of a sampled frequency table when expected marginal totals are known, Annals of Mathematical Statistics, 11, 427-444. 20 ©2013 BBM Canada Raking Ration Estimators • 21 ©2013 BBM Canada Raking Ratio Estimators Continued • Remarks on Raking Ratio Estimators • Always yield positive weights • May yield extremely large positive weights • It is possible to impose bounds (add constraints) on the weights In the case of BBM Canada: • TV Data is calibrated to: • Age-Sex, Reception Type, Household Size, Geography • Radio Data is calibrated to: • Age-Sex, Geography 22 ©2013 BBM Canada SORA Business Analytics Seminar DATA TYPES 23 ©2013 BBM Canada BBM Data is Big Data 24 ©2013 BBM Canada BBM Data is Longitudinal Data Longitudinal Data • • • • • • Panelists can remain in the panel for up to approximately a maximum of five years Data is granular to the minute (1,440 minutes per day) Data is captured 7 days a week Binary observation (1,0) for every encoded station Additional data appended to each observation to indicate if tuning was live vs. playback, in-home vs. out-of-home Each panelists answers a series of qualitative questions (industry, occupation, education, etc) 730 stations x 1,440 minutes x 365 days = up to 383.7 million observations per panelist per year!!!! 25 ©2013 BBM Canada BBM Data is Spatial Data Spatial Data • Every Panelists has an assigned FSA LDU which allows for a Latitudinal and Longitudinal coordinate • • • • 26 ©2013 BBM Canada Currently the FSA LDU of the panelists home is captured Future plans will also capture the FSA LDU of the panelist’s place of work (if applicable) This FSA LDU also maps to geo-demographic segmentation (Modified Environics Analytics PRIZM_C2) Data can also be overlaid with data from other sources such as Statistics Canada Census Data SORA Business Analytics Seminar BASIC OVERVIEW OF SPATIAL STATISTICS 27 ©2013 BBM Canada What is Spatial Data? Data which represents objects defined in a geometric space. This data can be one of many data types: • Points • Lines • Polygons • Surfaces 28 ©2013 BBM Canada What are Spatial Statistics? There are many types of spatial analysis including, but not limited to: • Spatial Data Analysis • e.g., Factor Analysis or Principal Component Analysis • Spatial Autocorrelation • The degree of dependency among observations in geographic space • Spatial Regression • Capturing of spatial dependency in regression analysis • Spatial Interaction • Estimation of flows between geographic locations • Spatial Interpolation • Estimation of the variables at unobserved locations • Spatial Diffusion Models • Estimation of spread through communications hubs 29 ©2013 BBM Canada The flow map of Napoleon’s ill-fated march on Moscow 30 ©2013 BBM Canada Spread of the Bubonic Plague in Europe 31 ©2013 BBM Canada Spatial Interpolation • Spatial Interpolation is useful for BBM Canada as it allows us to infer tuning/viewing for unobserved respondents in the market • There are many methods to interpolate data including: • • • • • 32 Kriging Inverse Distance Weighting Nearest Neighbour Cubic Splining Polynomial regression ©2013 BBM Canada SORA Business Analytics Seminar DATA-ViU: AN APPLICATION OF SPATIAL TECHNIQUES 33 ©2013 BBM Canada Data-ViU: Radio Listening Example Visualization • Radio Listening data from the Toronto’s PPM panel classified in quintiles • Overlay with census data or any other proprietary data • Any stored locational data from Google Earth Pro can be added • Example: Radio Format Group -Rock/Classic/Modern/ Alternative/Mainstream Rock/etc. (link to Google Earth Pro) Demo Web site 34 ©2013 BBM Canada