This Description of Methodology dated 10th July 2018 over-writes and replaces all earlier versions. It describes the measurement methods and processes utilized by Broadcast India 2018, including the most recent increase in the design sample to 44,000 panel HHs. The document provides details of edit and ascription rules used during production processing to maximize the quality and reliability of BARC published audience estimates.
The entire BARC India process can be broadly bucketed as follows:
Establishment Survey - A research study used to gather specific details of households and individuals to be used together with Census data in the preparation of universe estimates for TV audience characteristics – geographic, demographic, socio-economic status, etc. The Establishment Survey also serves as a randomly selected pool of TV owning households for use in the ongoing selection and recruitment of panel households
Listing Study- Supplements Establishment Survey in providing randomly selected pool of TV owning households for use in the ongoing selection and recruitment of panel households
Control Variables and Panel Design – Processes used to maintain panel that is representative of Universe
Panel Locations & Identification - Identification of a specific sample locations
Panel Selection and Training- Selection, recruitment, meter installation & training of household members
Panel Management- Supervision of panel operations with strict adherence to established standards
Measurement and Viewing Data Capture- In-home measurement technology used to identify & capture TV viewing events
Processing, Audience Estimation and Reporting- Process of error checking, pre-processing and attribution rules, validating, weighting, projecting to universe and delivering audience estimates to BARC India clients in a form suitable for reporting, analysis and commercial use
BARC India Media Workstation (BMW)- BARC’s desktop software application used to report and analyse audience data in the format required by individual customer segments
Given below is the overview of the high-level process flow:
The Universe Size of Total HHs, TV HHs, and their related demographic details was estimated basis the Universe Establishment Survey, Broadcast India 2018. The study was commissioned by BARC India and covered 300,000 households across the country. The study estimated the Total and TV population of households and individuals, their location, demographic distribution, connection type, language preferences and other media consumption. To ensure credibility of the survey respondent data, the Research Agency performed ground quality checks, telephonic checks and real time workforce tracking checks. The information collected from each survey respondent included the household head’s name, family member demographics, number of durables owned, education of the household’s chief wage earner and physical address. The Establishment Study (as well as the Panel) cover all parts of India except certain geographies that are unreachable due to harsh terrain, distance or areas with political unrest and safety concerns to field workers. These ‘uncovered’ geographies together account for approx. 0.6% of households and 0.7% of individuals as per the Census of India 2011.
Since these areas are not covered in the study, their estimated Universe Size is also excluded from the BARC India TV Universe Estimates.
Uncovered areas include Andaman & Nicobar Islands, Lakshadweep, Kashmir valley, Kargil, Ladakh and Arunachal Pradesh (except Itanagar).
The Universe size of all HHs and TV Owning HHs has been estimated using data obtained from Broadcast India 2016, applied to population growth projections from the Census of India projected through 1 March 2016.
To acquire additional sample locations/addresses for the remaining houses in geographic areas where the sample size was low and falling short based on the determinant variables, BARC India accessed data from ‘Listing Studies’ done in past 2 years. A total of ~100,000 HH were available with BARC India. Further shortfalls were met using geo-target based probability sampling technique.
Post rigorous statistical analyses, 9 variables were identified as having the highest impact on TV viewing, and are hence considered for representation of the TV universe (Control Variables). Of these, the Primary Control Variables considered are State, Town Class location of the household and the household’s socio-economic status as determined by the New Consumer Classification System (NCCS), an industry agreed upon standard. Secondary Control Variables considered either at panel composition or weighting are HH Size, Gender, Age Group, Languages Spoken at Home + Language Most Often Spoken at Home, Education of the Highest Educated Individual in the household and Mode of Signal Reception.
Respecting the order of the Government of India, effective with the April-June 2016 calendar quarter BARC initiated a process that phases out all Analog Cable HHs from the panel in all State Groups except TN/Pondicherry, with a completion deadline of having No Analog Cable panel HHs within any State Group except TN/Pondicherry from 1 July, 2017 onwards. Starting 1st November 2018, BARC India stoped measuring analog TV homes in TN/Pondicherry market. BARC India had stopped analog reporting in other markets except TN/Pondicherry in July 2017.With effect from 01st November 2018 BARC India will only measure and report data from digital TV homes in the country.
The NCCS classification of a household is based on two main variables:
1. Education of the household’s chief wage earner, defined as the person who contributes the most to payment of household expenses.
2. Household ownership of 11 specific durable goods. The 11 durables collectively owned by household members and considered in the NCCS classification of households in India are as follows:
a. Electricity Connection
b. Ceiling Fan
c. LPG Stove
d. Two Wheeler
e. Colour TV
g. Washing Machine
h. Personal Computer/ Laptop
j. Air Conditioner
k. Agricultural Land
There are 12 grades in the NCCS - A1, A2, A3, B1, B2, C1, C2, D1, D2, E1, E2, and E3. The table below shows the various NCCS categories, where A1 represents the highest class and E3 being the lowest.
The initial panel of 22,000 HHs was allocated per state group/metro based on RE concept. RE or the “Relative Error” concept is a type of statistical sampling error described as the deviation in percentage of the observed value from the actual/expected value from the selected sample. Based on learnings over the course of time, the RE methodology will be updated.
As the sample size is increasing (currently at 44,000 HHs), REs are naturally reducing. For the increased sample, BARC has also considered improved weighting efficiencies for designing the panel (Lesser difference between individual with lowest weight and individual with lowest weight = Higher Weighting Efficiency).
Statistical weighting allocations are done to ensure proper representation of the sample to the Universe. Detailed calculation of all panel control variables (much beyond the reporting cuts) was used to select the BARC India sample. The panel design considered each individual State, detailed Town Class and NCCS; along with other panel control variables - Languages Spoken at Home + Language Most Often Spoken at Home, Education of the Highest Educated Individual in the household and Mode of Signal Reception were used to design the sample. To illustrate - for designing the panel, the town classes were split into (1) Mega Cities (population 75L+), (2) 40-75L, (3) 10-40L, (4) 5-10L, (5) 2-5L, (6) 1-2L, (7) Below 1L Urban, (8) Above 5k Rural, (9) 2-5k Rural and (10) Below 2k Rural.
All cities with population above 5 lakhs as per Census 2011 were selected individually (except for Srinagar). Sample allocations for all other town classes within a State are based on the town-class group which is further allocated to PPS (Population Proportionate Sampling) among TV owning Households. Selection of actual Towns /Villages was performed by means of systematic random sampling after arranging available Towns / Villages in descending order basis their TV owning household populations.
To ensure against any convenience sampling on the field, ‘clusters’ (or groups) of eligible HHs are created by BARC based on panel control variables. All households in a single cluster are equally eligible to be recruited, and any single household is representative of the relevant cell that the cluster aims to fill. IDs of these HHs are fed into a central ID Master, which uploads the address and other relevant details of the household, along with a priority, to the mobile tablets used by assigned field executives. The field executives are expected to approach only the HHs which are in the cluster, and convince them to join the panel. Once one HH in a cluster agrees to be a panelist, the remaining un-attempted HHs move back into the main pool for future use, and HHs that are rejected / refused to be a panelist are churned out of the database.
The panel is statistically representative of the entire country. To make the reporting sample representative of the Universe, weights are applied to the reporting sample to conform it to Universe proportions and convert household and individual panelist viewing events to Universe Estimates of TV viewing (more details in Section ‘Processing, Audience Estimation and Reporting’ below).
The field recruiter goes to the HHD (household) sample location assigned by BARC India Measurement Science, explains the purpose of the BARC India TV Measurement Service and then seeks consent from the chief wage earner and householder for registering with BARC India. If the HHD is eligible (i.e. no disqualifications basis media/research affiliations of HHD members, adequate GSM wireless signal strength, agreement to incentives provided by BARC India, confirmation of compliance) the field recruiter asks the householder to provide specific household and household member information via a standardized panel recruiting questionnaire administered by the recruiter using a computer tablet app. Presently, fieldwork for panel recruitment and ongoing maintenance is handled by MDL as well as two independent agencies, hereinafter called the Panel Management Agency (PMA).
An integrated “Panel Management Software (PMS)” has replaced the android software applications “Recruitment Installation and Training Application” (RITA) and “Recruitment Installation Training and Maintenance Application” (RIMA). This links the mobile tablet that the field executives carry to the server, thereby enabling capture and transfer of panel Household details via the wireless cellular network directly to BARC India’s central office server. Automated validation checks in-built in the PMS enable many quality control checks to ensure panel health. This provides BARC India with a fully automated data collection process for use at all stages of the panel HH relationship.
Strict confidentiality is maintained at all steps of the panel recruitment, training and maintenance process. Ongoing hygiene checks are performed on PMA fieldwork by BARC India and its Quality Control and Analytics (DQA) partners.
The viewing behaviour of panel homes is reported to BARC India daily. The BARC India validation process analyzes household and individual viewership behaviors, highlighting behaviors considered to be outliers (at individual/household level). Based upon validation results, Measurement Science asks the PMA to perform coincidental checks on these homes either telephonically or physically. Certain suspicious outliers are also checked directly by BARC India – bypassing the PMA. BARC India also involves a separate vigilance agency to check on outliers that it considers highly suspicious. Non-compliance is categorised as a behavioural issue of the household or a technical issue with the meter. Any discrepancy in information is noted at this stage. If it is a behavioural issue, the household is then re-trained. If non-compliance continues, then the panel home is dropped. If there is a technical issue with the meter, then the issue is resolved by the BARC India field and technical teams. Where needed, technical issues are raised with meter technology providers.
Panelist training and compliance maintenance are priority issues for the PMA. Pursuant to BARC India policy, those households that exhibit substandard compliance, when compared to BARC India standards, are retrained. If, after retraining, a household continues to underperform, it will be churned out of the panel.
The training protocol specifies two post installation training visits. The first visit is generally made 3-5 days post installation and includes training HH members in persons viewing button pressing, observing the working condition of equipment, verifying that the user manual is provided and available for use, etc. The second visit, generally made 10-12 days post installation, includes coincidental checks – whether TV is ON or OFF, channel viewed and persons viewing with retraining as needed in button pressing and confirming that the family member button assignments are correct. The PMS application also has a pre-loaded training module for this purpose.
To ensure up-to-date and correct household data as well as for periodic re-training, the Field SOP mandates a complete demographic check every six months of each panel HH.
A policy of systematically rotating panel homes is prescribed by BARC India policy, ensuring that panel secrecy is maintained, precluding long term panelist fatigue and allowing the installed panel sample of TV households to reflect changes that are occurring over time in the universe of TV households. Beginning with the second year, after the initial panel installation has been completed and the panel is firmly established, a four (4) year churn/forced turnover cycle is initiated. Under this procedure, the total India sample, including open clusters, is randomly divided into four exhaustive and unique replicates, each replicate representing 25% of all clusters in the current sample. Each forced churn sample replicate is assigned one of four consecutive 12 month periods (i.e. 4 replicates over 4 years equals the requisite 48-month sample replacement cycle) during which all panel households within that replicate will be de-installed and replaced with households randomly selected from a new replacement sample replicate. Ideally, every month approximate 2.0-2.1% of panelists will be replaced as a result of the normal and forced churn and sample replacement policy. After the initial 48-month period, the sample will be churned on a FIFO (First In-First Out) basis, ensuring adequate replacement before a household is churned out.
To ensure the reporting sample will not suffer from forced churn, replacement sample households are selected, trained and monitored for a minimum period of two weeks after which the replacement household is activated and the original household is de-installed.
To respect privacy concerns of panel households, the maintenance supervisors hired by the PMA are assigned responsibility for a specific group of individual panel households. Earlier in the development of the BARC India panel, different PMA personnel would visit a household for recruiting, installation, training and maintenance purposes. This practice made it difficult for BARC India or PMA management to assign accountability in situations where multiple field recruiters were visiting the same panel home. In the current situation, the PMA hires Maintenance Supervisors (MS) across its office locations and assigns each MS with responsibility for a specific group of BARC India sample locations. Each MS is responsible for recruiting, installing, training and maintaining the panel relationship with their assigned sample locations, ensuring proper compliance and provisioning BARC India with household and household member TV viewing data.
BARC uses TV set metering technology that captures watermarks embedded in the audio transmission of TV channel transmissions to identify the channel being viewed.
This technology captures TV usage, TV station identification and individual viewing through the use of two digital devices, one installed by the broadcaster (Embedder) at station head end/transmission site(s) and the other device, referred to as the “BAR-O-Meter”, that is installed on each TV set in the panel household.
Embedder equipment is placed at the Broadcaster’s headend where the Channel signal transmission begins. The device embeds a unique watermarked code in the audio component of the program content workflow. This code consists of the Channel ID & the time stamp. Each channel has its own unique code (or codes, in case the channel has taken a back-up). Once the unique watermark IDs are generated and assigned to each broadcast station cooperating with BARC India, the embedder is installed at the broadcaster’s headend transmission site and a special station specific electronic card is inserted. The results in the embedder continuously placing a time stamped channel name and watermark ID in the station’s content workflow. The watermark is an inaudible audio code made available to TV broadcasters that subscribe to and support the BARC India measurement of TV audiences. A master list of TV Station Watermarked IDs is stored on the BARC server and downloaded to BAR-O-Meters for the identification and measurement of TV Station viewing.
After receiving householder consent, the PMA recruiter connects the BAR-O-Meter to the TV set and establish a wireless network connection with the BARC server. The connection process automatically creates and assigns a unique HHID for that household. Once an ID is created for a HH, it is unique to that HH and is never re-used or assigned to another household.
Each meter system consists of a main unit, a display unit and probes that for BAR-O-Meters capture the audio output of the TV set.
Each main unit is equipped with a microprocessor and a modem. The main unit is placed near the TV set being measured in the panel household. Each main unit has a probe attached to it that is either placed near the TV set or connected to the line or audio out of the TV. The probe capture the identity of each tuned TV signal and feeds this information to the main unit where it is time stamped and stored for transmission as viewing events to BARC central site collection servers assigned to BAR-O-Meter measurement systems.
The method of individual person’s viewer identification for the BAR-O-Meter is a button pushing remote handheld device. The measurement system provides one handheld remote control unit for each metered TV set. The handheld device has buttons made available for assignment to household members who are asked as part of their panel participation to press their assigned button when they are viewing TV. Each panel household member aged 2 years and older is assigned a button on the remote control handheld unit. Separate buttons on the remote handheld device are reserved for use by guests, entering their gender and age bracket when viewing TV.
The TV set metering systems continuously and passively captures TV viewing events in real time, recording the time and duration of channel tuning events and capturing the viewership events of individual members ages 2+ that have pressed their viewer ID button to confirm their presence in the audience.
The main unit stores the individual time stamped events in memory for transmission to the BARC server at predetermined intervals throughout the viewing day. The BAR-O-Meter TV viewing event data are then received by BARC collection server where collected TV event data are simultaneously backed up and made available to pre-processing software.
The data from the collection server is first pre-processed, where errors and inconsistencies that may creep in due to technical issues are cleaned up. In this state, attribution rules are applied.
Data collected from the meters is in seconds. However, in keeping with international standards, all validations rules are on viewing sessions (blocks of time of TV Set on in the HH – Tuning; and of each individual viewing TV - Viewing) and reported data is in clock minutes. Hence, all data needs to be converted to clock minutes (i.e. HH:MM format, e.g. 12:00:00 to 12:01:00, 12:01:00 to 12:02:00 and so on).
Attribution rules are applied on the statement file at the pre-validation stage, i.e. after the data is received from collection servers for production processing and validation.
There are five conditions under which viewing behaviour is to be attributed-
TV Set Session
Magnetisation, i.e. linking viewers to the time the TV is switched on (at the beginning of viewing time)
Bridging (required only for BAR-O-Meters where no watermarks are present between two watermarked channel viewing sessions)
Individual Viewing Sessions within a clock minute
Channel Viewing Sessions within a clock minute
TV Set Session: Refers to the time the TV is switched on and off. If a TV set is on for 30 seconds or more in a clock minute, it is attributed as being on for the entire clock minute. In the BAR-O-Meter measurement system TV On and Off status is determined by the presence or absence of a watermarked channel. Since viewership of non-watermarked channels is not captured by the BAR-O-Meter, any viewing of non-watermarked channels is considered as TV Off.
Magnetisation: There is generally a gap between the time viewers switch on the TV set, move to the channel intended to be viewed, and press their viewing buttons on the BARC India remote. Unless removed, this gap would depress viewing by the duration from the time the TV is switched on and the individual button is pressed. A Magnetisation algorithm is applied in such cases and viewership of these individuals is ‘magnetised’ or linked back to the time when the TV set was switched on.
Bridging: Bridging applies only to TV sets measured with BAR-O-Meters for use when people put the TV set on mute for short durations. Unless this gap is ‘bridged’, it would be considered as TV off and the time spent viewing during the gap would not be captured. In order to include the gap as viewing time a bridging algorithm is applied when no watermark is present in between two watermarked channels for a certain maximum duration. For bridging, the following rules are applied on the TV set-
In case the channel before and after the non-watermarked duration is the same, viewing duration of the non-watermarked period is attributed entirely to this channel
In case the channels before and after the non-watermarked duration are different, viewing duration of the non-watermarked period is attributed alternately to the earlier and later channel, i.e. the viewing is attributed to the channel being viewed before the non-watermarked duration in the first, third, fifth (and so on) instances observed in the system; and the viewing is attributed to the channel being viewed after the non-watermarked duration in the second, fourth, sixth (and so on) instances observed in the system
Individual Viewing Sessions within a clock minute: There are rules applied to the second by second events that attribute viewing to one and only one TV channel for an entire clock minute. In each systems only one channel is eligible to receive viewing credit for each clock minute throughout the viewing day. if an individual is viewing a TV channel for 30 seconds or more in a clock minute, the rules are straightforward and viewing is attributed to that channel for the entire clock minute.
The rules become more complex when viewing during a clock minute involves multiple channels for a total of 30 or more seconds additional rules are required as described below for processing BAR-O-Meter event data.
Channel Viewing Sessions within a clock minute: Individuals can view multiple channels within a single clock minute. However, only one channel will be assigned the viewing in each clock minute. To assign this viewing, the following rules are applied.
Rule 1 - Only one channel watched: The viewing for the entire clock minute gets attributed to that channel
Rule 2 - Multiple channels watched with different viewing durations: Viewing is attributed to the channel with the maximum viewing duration.
Rule 3 - Multiple channels watched with two or more channels having the same maximum viewing duration: There are two scenarios for this rule-
Scenario 3a – One of the channels with the maximum viewing duration moves into the next clock minute. In this case, viewing is attributed to the channel moving into the next clock minute.
Scenario 3b – None of the channels with the maximum viewing duration moves into the next clock minute. In this case, viewing is attributed to a random channel from among those channels having the maximum viewing duration, using a random allotment algorithm.
It is pertinent to note that the 30 seconds or more rule, wherever applied, refers to a total of 30 seconds in a clock minute – whether consecutive or not.
The preprocessed minute level data is then subjected to further processing with software that performs data validation and weighing.
Validation of viewership data is a daily process performed at two levels – Validation rules (called ‘Metarules’) that validate the data for identification of statistical outliers; and Screening rules, a set of more stringent rules, applied to channels that have been confirmed as having attempted tampering of panel households following a rigorous process of Vigilance investigations and raw data analyses by Data Scientists. Metarules and Screening Rules consist of documented and strictly controlled rules applied in a transparent and systematic manner during daily production processing.
Validation rules and their application are subjected to external audit but not otherwise disclosed to prevent individuals who might attempt to tamper with panel HHs from gaining valuable insights. Those households and individuals that fail Metarules and/or Screening rules are identified as outliers and removed from the reporting data of the day.
The Weighting process assigns a weight or factor to each household and each household member that reflects their proportionate representation of the universe. Sample weighting is a statistical process used to compensate for imbalances that may exist between the daily In-tab reporting sample of BARC panel households/individuals and the estimated universe of Indian TV households/individuals published by BARC. Weights are applied separately at Individual and Household levels. The twin data sources used by BARC in preparing Annual Universe Estimates are Census 2011 updated to reflect the current year’s population and BARC’s Broadcast India Establishment Survey that defines households with a TV set by state and TC together with their NCCS profile.
The final weighted and projected audience viewing output is encrypted and made available to BARC India subscriber through the BARC India Media Workstation (BMW) desktop software; a powerful reporting and analysis platform developed for BARC India to present India TV audience data in a form suitable for use by subscriber media companies, media agencies and advertisers.
On a daily basis, BARC India monitors the output of preprocessing, processing as measured by the daily In-tab for key sample strata. The In-tab is defined as the number of HHs and Individuals in the final output. Fluctuations in the In-tab level are typically the result of new HH recruitments, panel churn, weather and environmental events that impacts viewership, GSM network problems that affect data collection and changes in household and household member viewing behavior. As previously mentioned, statistical weighting of the In-tab sample is a best practices method of compensating daily variations in the In-tab sample composition.
From a reporting perspective, BARC reports the following-
State Groups x Town Class: 39 markets. The 6 Mega Cities are reported individually; 10-75L towns and <10L Urban towns are reported separately in 5 State Groups while all Urban is clubbed in the other State Groups; and Rural is reported as a single unit in each state.
Age Groups: 2-14 yrs, 15-21 yrs, 22-30 yrs, 31-40 yrs, 41-50 yrs, 51-60 yrs, 61 yrs+
NCCS Groups: NCCS A, NCCS B, NCCS ABC and NCCS CDE.
BARC India has contracted an independent company as its partner to collate the play-out data aired by cooperating BARC India broadcasters, ensuring correct integration of the broadcast data stream thrown out by the embedder at each TV Station and the watermarked time stamped BAR-O-Meter viewing events. Currently more than 500 channels are being monitored. The monitored content is downlinked and continuously monitored for each BARC India watermarked channel. Automatic content recognition software (“ACR”) technology also known as audio signature or audio fingerprinting technology is used to identify programs, promotions and ads.
BMW is BARC India’s licensed desktop software distributed to BARC India media, agency and advertiser stakeholders for their use in audience reporting, analyzing, targeting and planning. Each Thursday, on a regular weekly basis, BARC India issues daily reports of TV audience viewing using the BMW tool. The BMW software tool enables the end user client to access customized reports, graphical representation, individual analysis, audience movement and other features that make it easier for users to measure performance and establish trends. Some of the key features of BMW which are accessible to the client include:
Analysing viewership based on time band based on the pre-identified target group
Deeper advertising led analysis by identifying and scheduling ad-spots
Automated process of media planning (Ad budgets/reach/frequency/CPRs)
Multi-channel viewing pattern & rating analysis
· Classic Grid analysis:
It relates both to time-bands and to programs, builds a grid of programs, filtered by time slots, with the daily audience values (audience variables) for each program.
· Rating Curve: BMW uses a more flexible module, which allows making multidimensional selections and displaying more than one audience variable at the same time.
Separately BARC has published documents that describe in detail the statistical limitations of BARC TV Audience Measurements and policies regarding the permissible use and public disclosure of BARC data.
BARC Out-Of-Home Audience Estimate Model
Description of Methodology
This document provides and overview of the entire BARC India process for calculating out-of-home (OOH) audience estimates (Figure 1). The document is separated into the following eight sections.
1. OOH Universe Estimate (UE) Study for estimating the incidence of individuals visiting eateries in urban India;
2. OOH panel design;
3. OOH panel recruitment and installation processes;
4. Footfall measurement;
5. Identifying individuals from TV panel eligible for OOH;
6. OOH algorithms:
a. Footfall extrapolation;
b. Clustering of Individuals from footfall measurement;
c. Attribution of demographic variables; and
d. Fusion of OOH viewership to the TV panel.
7. Reporting and data conversion; and
8. Limitations of OOH audience measurement.
Figure 1 . Process for generating OOH audience estimates.
OOH Universe Estimate Study for estimating the incidence of individuals visiting eateries in urban India:
An estimate of the size of universe (UE)for individuals visiting eateries and their related demographics was produced through a monthly rolling primary research study. The scope of this survey consisted of individuals residing in urban India across all 16 state groups reported by BARC. The sample design was based on the projected population from Broadcast India 2018. Through this study, the total number of individuals on daily basis was estimated. Estimates were based upon:
the preferred days of visiting eateries;
the frequency of visiting eateries;
the regularity of the previous visited eateries;
the type of eateries visited; and
the genres watched at eateries.
Data from the survey was calibrated and projected. Calibration was conducted through cell weighting aligning to TV / Non-TV X State group X Town Class X NCCS X Age Group X Sex levels Data was projected to the population derived from Broadcast India 2018.
OOH footfall is dynamic and differs between weekdays and weekends. Therefore, survey responses were attributed to the day of the week. This attribution occurred at 4 different levels: (a) Monday to Thursday; (b) Friday; (c) Saturday; and (d) Sunday. These levels formed the basis for the most preferred day for visiting eateries – this effectively producing different OOH estimates on daily basis.
OOH panel design
In order to draw a probability sample a sampling frame – which is a complete list of statistical units covering the target population – is required. BARC India used a pre-existing sampling frame for drawing sample. This sampling frame was a comprehensive list of all eateries along multiple variables such as state, Town Class, Restaurant type, TV availability, Town Name, Pin-code. The frame was built by combining multiple databases of eateries sourced through multiple agencies. The most exhaustive database at city level was used as the primary database and data from the other databases were added creating the larger detailed database of eateries.
The OOH panel comprises of 1,050 eateries. Statistical stratification was conducted in the allocation of sample, ensuring proper representation. Stratification was across the following variables:
Restaurant type; and
All towns with a population greater than 20 Lakhs were included in the OOH Panel. The remaining medium and small size towns were selected using systematic random sampling. Population Proportionate Sampling (PPS) was then used for sample allocation amongst the remaining towns.
In order to prevent convenience sampling occurring on the field, ‘clusters’ (or groups) of eligible eateries were created based on panel control variables such as type of eatery and pin-code. All eateries in a single cluster are equally eligible to be recruited, and any single eatery is representative of the relevant cell that the cluster aims to fill.
OOH Panel Recruitment and Installation
Using the location assigned by BARC India, a field recruiter visits the eatery and explains the purpose of the BARC India OOH TV Measurement Service and seeks consent from the Manager / Owner of the eatery for participation. If the eatery is eligible (i.e., name of eatery is correct, type of eatery is matching and the owned TV sets are in working condition), the field recruiter collects specific information for the eatery such as operating hours, seating capacity followed by other recruitment information. The recruitment questionnaire is administered via a mobile phone or tablet. All fieldwork for the OOH panel is managed by Meterology Data Ltd. (MDL) and their sub-contracted panel management agencies (PMA).
An integrated “Panel Management Software” (PMS) links the mobile tablet that the field executives carry to the server, thereby enabling capture and transfer of OOH panel details via the wireless cellular network directly to BARC India’s central office server. Automated validation checks in-built in the PMS enable many quality control checks to ensure OOH panel health. This provides BARC India with a fully automated data collection process for use at all stages of the OOH panel relationship.
Footfall measurement is necessary in order to report daily reach and impressions. Footfall is measured in almost 200+ eateries across multiple cities and mapped to the viewership of 1,050 eateries.
Eatery footfall is measured manually using a first-in/first-out (FIFO) technique. Footfall is measured in-person using a device in which the ticker APP is installed. This process then captures individuals going in or out of the unit. This manual approach is the most common and reliable type of footfall measurement allowing for accurate footfall analytics. Footfall measured and analyzed in this way are very rich in terms of the data provided (people counts, dwell times, movement, heat maps, time between visits and more).
Footfall was captured throughout the operating hours of the eateries across the week within each State group X Town Class. Details captured were as follows
Count of patrons entering and exiting;
Sex of each patron; and
In-time and out time of patrons.
A schedule of eateries is prepared for everyday of the month thereby allowing for the rotation of measured eateries – ensuring adequate spread of the eateries throughout the year.
Unbiased Validation rules and Quality checks were set-up to ensure there are adequate footfalls being captured from data.
Identifying potential individuals from TV panel for OOH
BARC measures in-home (IH) and OOH television viewing of individuals who reside in a household with at least one TV. Therefore, in order to correctly attribute OOH viewing to the BARC TV panel, the proportion of OOH viewing coming from individuals residing in TV households needs to be estimated. This step is done through a multi-stage modelling system. This system no only allowed for an understanding of the proportion of OOH viewing which should be attributed to the panel, but also, the correct individuals in the panel which could be eligible to receive OOH viewing.
In order to identify eligible panel individuals for OOH viewing a survey was conducted amongst members of TV panel from Urban India. The goal of the survey was to gauge the incidence and behavior of these individuals for: (a) likelihood of visiting eateries on daily basis; (b)preferred days of visiting eateries; (c) frequency of visiting eateries; (d) last visit at an eatery; and (e) type of eateries visited.
Using the data collected from survey a likelihood (i.e., probability score) was computed for each individual visiting eateries from TV panel. Higher the score for the day, higher was opportunity for individual to be the most probable individual to receive OOH viewership.
Multi-stage data modelling techniques were adopted to produce TV+OOH Viewership results in BMW. The were 5 different models used.
Despite OOH viewership being measured in 1,050 eateries with roughly 1,500 meters, in any given day, the maximum eateries in which footfall gets measured was measured is 200. Hence, a 5-tier hierarchy approach for attributing footfall from eateries to eateries without footfall was required. This was done in order to find out the best fit eatery to map the footfall and viewership. The hierarchy of the approach is as follows:
1. Eatery type;
2. City / town;
3. Town Class;
4. Seating Capacity; and
5. State Group.
When the number of TV sets in an establishment was more than than one, the footfall was divided between TV sets using Bayesian probability. This ensures that all TV sets in the eatery are directly mapped to specific viewership and footfall data.
Clusters were created using a hierarchical clustering technique for clubbing the viewing sessions at the channel level. The hierarchy was based on the following variables:
In time for patrons;
Post creating clusters, in-time and out time was averaged using the start- and end-time and number of all individuals appearing in that clusters. A new Cluster was created every 30 minutes ensuring zero duplication of channels within same cluster.
Usable clusters were identified basis top channels by duration in each cluster and the count of individuals from HH TV panel to which OOH Clusters needed to be allocated in TV panel. Details of computing count is covered in last section of Algorithm
Attribution of Demographic Variables
The variables captured in the OOH individual UE survey data provided distribution patterns of Age group X NCCS for each genre watched within eateries. This distribution proportion of Age X NCCS for each genre from the study was attributed to OOH clusters at lowest possible level. This allowed for a channel mapping with Age X Sex X NCCS for each usable cluster in each State X Town Class.
The weighting process assigns a weight or factor to each Cell that reflects their proportionate representation of the universe. Sample weighting is a statistical process used to correct for imbalances that may exist within the realized sample. Weighting occurs on a daily basis and assigns OOH individual weights which are then applied to OOH clusters.
Data output from the OOH Monthly rolling UE is used for OOH estimation and weighting purposes.
Weighting was done using a cell weighting technique. The variables used for weighting were as follows:
o Reported state group (16 levels);
o Reported Town Class within each state group (up to 3 levels)
o Reported NCCS (3 levels);
o Sex (2 levels); and
o Reported age group (7 levels).
Applying OOH Viewership to TV panellist
The count of the individuals from the TV Panel to which OOH clusters have to be applied were computed dividing the “OOH UE weight” by “weighting factor of TV Individual” at the lowest common weighting levels.
Using the probability Scoring system (explained above), the individuals of TV panel were sorted in descending order of Score and best fit procedure was used to ensure OOH clusters were properly allocated. This allocation was accomplished at the State group X Town Class X NCC X Sex X Age group level. This means that only a subset of individuals from TV panel carry TV and OOH viewership data for reporting TV+ OOH Viewership
Reporting and Data conversion
The final data is converted to a usable and reporting software readable format using BARC’s BMW convertor. Post this conversion, all channels are masked and grouped at 2 levels for In-Home + Out-of-Home reporting:
Star group Channels: Individual channels will be visible to Star and Media agencies
Non-Star channels: grouped under ‘Other’.
All existing BMW metrics will be visible for In-Home TV (i.e. TV) and In-Home TV +Out-of-Home TV (i.e. TV+OOH TV) for clients.
Limitation of OOH measurement
Like any other form of measurement employing samples, OOH audience estimates are subject to various forms of errors. The following lists some key limitations with regards to the OOH audience estimates.
Since only a portion of the population is observed, audience estimates are subject to sampling error. Users are encouraged to consider relative errors in relation to audience estimates.
Relative errors increase for data points at finer cuts.
Due to possible errors introduced through modelling, modelling error may be present and these errors may differ by channel.
‘In-Home + Out-of-Home’ and ‘In-Home’ reports will be available in BMW; however, ‘Out-of-Home’ only reports will not be available.
Due to variability in viewership by channel, the impact of the addition of OOH viewing will differ by channel. Since OOH viewership is only being added to Star stations within Urban India, only these stations will have additional OOH viewership. Rest of the stations are clubbed under “OOH others” in BMW. ‘In-Home + Out-of-Home’ audiences will be the same as ‘In-Home’ audiences for all other stations or for Rural India audience estimates.
Final Data is subject to modelling accuracy.
The Universe size and composition of OOH will change once a month and hence may introduce variability into audience estimates.
Footfall is measured manually using FIFO method therefore ATS as a measure for only OOH may not be appropriate