Genauer Titel der Studie

Video Impact Study Retail

Jahr der Veröffentlichung bzw. des Studien-Starts

Veröffentlichung: 2019 / Studienstart 2018


Online/Digital, TV

Kontaktperson für Rückfragen

Torsten Müller-Klockmann

E-Mail-Adresse der Kontaktperson

Urheber / Vermarkter

GfK / Facebook

1. Zielsetzung der Studie (inkl. Forschungsfragen)

Evaluation of efficiency and effectiveness of Facebook and Instagram video ads for grocery advertisers and comparison in a cross media (TV) context

2. Untersuchte / gemessene KPIs (bitte mit kurzer Erläuterung)

- How do ads on the Facebook family of ads impact the purchase probability of reached consumers and what can Facebook add to the impact of TV?
- What are differences between campaigns advertising the retailer in general vs campaigns advertising for a specific product category?
- Are viewtimes of ads correlated with sales lift?
- What are short term ROIs of the ads at given reach rates?
- What are typical incremental reaches and reach overlaps between TV and the Facebook family of apps at given reach rates?

3. In der Studie untersuchte Medien (intramedial / intermedial)

Facebook / Instagram

4. Studien-Konzept (Beschreibung in wenigen Sätzen)

We ran 5 Video campaigns (min 60% share of ad impressions were video ads >6s) on both Facebook/Instagram and TV for grocery advertisers and GfK calculated individual models for sales impact in their GXL panel for each of them.
Calculation of individual effectiveness models are based on logistic regressions for each of the 5 campaigns.
Campaigns had different runtimes within the period of 01/2018 to 12/2018.
Facebook budgets ranged from 111 – 170 TEUR, TV between 1.9 and 22.8M EUR (gross)
Campaigns targeted broad audiences reaching 4M - 14M people on the FB family of apps and 32M-58M people on TV.

5. Auftraggeber


6. Institut(e) / Subinstitute


7. Eingesetzte Methoden ( Multi-Methoden-Ansatz oder Single-Ansatz)

Single source measurement (with partial data imputation) of purchase behavior and media contacts in a panel

8. Art der Veröffentlichung (Broschüre, Tabellen-Band, zählbarer Datensatz)

Presentation deck, one pagers

9. Grundgesamtheit

Total population 14+

10. Erhebungsmethode (inkl. Technische Messung)

Technical device measurement in a single source panel (GXL):
Purchase behavior is measured via household scanner as part of GfK’s consumer panel.
Facebook/Instagram ad contacts on all devices are measured through panel data link with Facebook, TV ad contacts are measured via TV meters for all devices in panel households for 9000/19000 panelists and data imputation on TV usage for a subsample of 10000/19000 panelists who are not measured via a TV metering device.
Explanation of TV measurement:
- Panel members are equipped with mobile communication-based TV advertising device
- Advice is to place the phone next to the TV remote control
- Every person in household must log on/off with the mobile phone
- TV consumption from 06am-01am for 19 advertising relevant channels is covered
- No continuous measurement, but 8 seconds sequences are recorded: 3 times per minute from 4pm to 10pm, 2 times from 10am to 3pm and once from 6am to 9am as well as from 11pm to 0am
- The audio data is stored on the mobile measurement device and transmitted to GfK overnight
- Ebiquity provides reference audio files
- The audio sound matching technology identifies the watched channels
- The 8 second sequences are completed by the convention that a match is valid until the next match

11. Feldzeiten / Untersuchungszeitraum

All measured Facebook campaigns were live between January 2018 and December 2018

12. Fallzahlen (ungewichtet) / Wellen

Online: ~19,000 individuals with 70% also reporting on FMCG purchase behaviour.
TV (as of H2 2017): 9,000 individuals 14+) as a subsample of the online panels + 650 individuals 14+ who are not online.

13. Stichproben-Auswahl (z.B. Quote, Random, Einsatz von Access-Panels)

Quota sample drawn from the GfK Consumer Scan panel, TravelScope & MediaScope.
Weighting based on data of the Federal Statistical Office / ma Radio.
People from German-speaking households aged 6+ years (currently reported: 14+ years).

14. Ggf. Rekrutierung / Incentivierung der Probanden / Interview-Dauer

The study is based on the continuous GfK GXL measurement, so no additional recruitment/incentivation for this study took place.

In general, households that report within the GfK Consumer panel participate in a bonus program, where they collect points based on their reporting behavior.
The households can redeem the points for various premiums, such as a coffee machine etc. Households that additionally report their media usage behavior within the GXL benefit from additional points.
Recruitment details:
- The GXL is recruited from existing GfK panel households.
- Households that are part of the ConsumerScan panel and do report frequently over a longer period of
time are recruited for the online desktop/ laptop measurement and/or TV measurement.
- These panelists are also recruited for smartphone measurements.
Over-fulfillment or under-fulfillment of quotas
- The recruitment of all three samples (desktop/laptop, TV and mobile) derives cells that are underfulfilled and cells that are over-fulfilled in proportion. At present, young people leading a household are regularly underrepresented. Generally, it would be possible to allocate the sample by excluding households, respectively mobile measured individuals, from over-fulfilled cells within the GXL.
- To optimize the sample approach, all households/individuals are included, but overrepresented cells
are given lower weighting, while underrepresented cells are given higher weighting.
In doing this, GfK uses a statstically proven weigthing method.

15. Verwendete Datenquellen (z.B. Werbeaufwendungen von Nielsen Media Research)

TV detailed media schedule – source: Ebiquity
TV reference numbers (reach, GRP, frequency) – source: AGF

16. Messung / Berechnung der Kontakte bzw. Kontaktwahrscheinlichkeiten

Technical measurement / data imputation:
Facebook/Instagram ad contacts on all devices are measured panel data link with Facebook, TV ad contacts are measured via TV meters for all devices in panel households for 9000/19000 panelists and data imputation on TV usage for a subsample of 10000/19000 panelists who are not measured via a TV metering device.

17. Eingesetzte Analysetechniken

Statistical technique: Binary logistic regression
Control variables: loyalty classes, loyalty index (expenditures of the advertised product), promotion (weekly promo share), demographics (household size, household income, gender, age)

18. Gewichtung (vor oder nach Analyse, Quelle der Gewichtungsvorgaben)

Weighting procedure Linear Weighting
- The GfK linear weighting has replaced the classically used method of iterative proportional fitting.
- Linear weighting is more flexible regarding variables and its combinations as well as more efficient and more accurate.

Cell plan
The cell plan contains demographic target specifications as well as potentials of online media users.
- Demographic variables are: Federal state, Region X Gender, Age (in steps of 10) X Region, Gender, Age (in steps of 10), Household Size (for 1 person households crossed with age), Household Size X Region, Age X Education.
- Online Potentials are: Tablet X Gender, Tablet X Age (in steps of 10), Tablet X Education, Smartphone X Gender, Smartphone X Age (in steps of 10), Smartphone X Education, Smartphone IOS, Smartphone Android, Tablet IOS, Tablet Android.
- Weighting requirements for the TV reception type is missing in the TV cell plan as this is not registered within the GXL. A survey would not deliver sufficiently accurate results. As only 19 channels are measured, an estimation of the channel reception is also not possible. Due to economical reasons, no in home research is carried out.

Source of target specifications
Demographics: Federal statistical office: micro census
Online total mass: 2016 Establishment Survey by GfK. (Since 2017: ma Radio)

19. Ggf. eingesetzte Fusionstechniken

For ex post advertising effectiveness analysis only TV contacts are imputed --> PMM: Predictice Mean Matching
In order to make full use of all measured online panelists, the TV measured panelists are imputed into the bigger online sample (from 9,000 panelists to 19,000 panelists).
„Predictive Mean Matching“ is an approach for data completion. The missing values here are the not existing TV contacts within the total mass of non TV measured panelists.
1. Conditional predictive mean: in a first step, a multiple regression is assembled to the available contacts on the basis of the link variables.
2. Connecting donors with recipients: in a second step, for each recipient the best possible donor is searched on basis of the regression.
3. Translating the contacts: in a third step, the TV contacts from a donor are transferred to a recipient with no TV measurement.

Variables for Predictive Mean Matching:
Household Size
Locality Classification
Area of residence
Family/Living environment
Social Status
Occupation (employee, official, worker, retired person…)
Employment level (part-time, full-time…)
Social class
Class of prosperity OECD
Specific Variables:
Queried TV consumption by time and channels
Measured online usage duration during the campaign

20. Verwendung von Benchmarks (bitte Erläuterung dazu, z. B. inter- oder intramedial, nach Branchen, nach Werbeformen etc.; Quelle von externen Benchmarks, Art der Berechnung)