How do we ensure GWI Core is representative?
To ensure our Core sample is representative of the online population aged 16-64 in each country, we set quotas on age, gender and educational attainment. In some countries, we use additional criteria such as race and ethnicity (read on for more detail).
We update our quotas at least once a year using a range of international and national sources including the World Bank, the International Labour Organization, the UN Population Division, Eurostat, government departments and national statistics sources. The latest available figures from these sources will typically be 1-2 years old, so we examine long-term trends and generate forecasts for the current year.
Once our quotas are set, we share them with our panel partners, who then proceed to recruit participants accordingly to achieve a demographically representative sample. When fieldwork begins, some groups might respond faster than others. In such cases, in order to ensure the timely release of data, we may “relax” certain quotas and then weight the sample so that each group is still represented fairly. For instance, we might have required 1000 respondents overall, with 500 being men and 500 being women. If we find that we are receiving more responses from women than men, we might accept a final composition of 475 males and 525 females, and then weight the sample to ensure that each gender accounts for 50%.
Typically, we find the hardest groups to survey are younger men in mature markets and older women in fast-growth markets. Nonetheless, our panel partners work constantly to ensure that their panels are as representative as possible.
After fieldwork, we assign a “weight” to every respondent based on their age, gender and education profile. As with quotas, in some countries, we use additional criteria (read on for more detail). Regardless of the criteria used, this process allows us to calculate approximately how many real world internet users are represented by each respondent. For example, we might know that a response by a female respondent aged 16-24 represents 15,000 similar individuals in a given market. So, each time that respondent selects an option, the corresponding audience size for that option is increased by 15,000.
The average weight a respondent receives varies by market, and is largely influenced by the size of the population in that country, as well as the ease of conducting research there. We're most representative in countries with smaller populations, such as Ireland, Singapore and New Zealand. In these places, a Core respondent will typically represent about 1,000-3,000 internet users. Meanwhile, in countries with particularly large populations, such as China and India, or countries where respondents are harder to recruit, such as Nigeria, one respondent will typically represent 30,000-70,000 internet users.
Why don’t we use other criteria for our quotas and weights?
Some studies set quotas on other criteria, such as income, region and ethnicity. However, such information about a country’s online population isn’t always readily available or accurate. For this reason, and to maintain a harmonized approach, we stick to age, gender and education in most countries.
For the most part, our age, gender and educational attainment quotas allow us to recruit respondents from a good mix of backgrounds through natural fallout (jump to the Local regions section for an example of this). However, when the make-up of the population means that additional quotas are particularly important for understanding audiences, we do make some exceptions.
The most notable exception we make is in the USA, where we set quotas on age, gender, race/ethnicity, income and region. This applies both to GWI USA and GWI Core. Additionally, in the UAE, we set quotas on nationality, while in Saudi Arabia we developed a specific weighting framework which includes interlocking age, gender and nationality quotas.
In all Core markets, we ask respondents which region they currently reside in. Although we don’t set quotas on region, the natural fallout of respondents aligns very closely with what we know about the population distribution in each market.
For example, the below comparison shows the distribution of our Core sample in the UK alongside data from the 2015 UK Census for individuals aged 16-64:
|Yorkshire and the Humber||8%||8%|
|East of England||9%||8%|
As the Core sample represents internet users aged 16-64, regions with a higher internet penetration will generally contain a higher proportion of the total sample (even in markets like the UK where total internet penetration is over 90%). Moreover, it’s also possible that our respondents will have a different interpretation of the region in which they live than the one assigned to them by the census based on their postcode, particularly if they live on the border between two regions or near a large city.
In order to keep the survey as short as possible for respondents, we don’t currently collect county or city level data other than in the USA and China.
In all Core markets, we ask respondents about their annual household income. However, because income can be a sensitive matter for some, respondents aren’t required to provide an answer.
For this reason, in most Core markets, we use education quotas instead of income, distinguishing between those who've achieved primary, secondary or tertiary level education. An additional benefit of this approach is that education tends to be a more stable metric than household or personal income, which can be subject to short-term fluctuations.
As per the previous example of local regions, we find that our age, gender and education quotas provide an accurate view of the composition of the online population, with income falling out naturally.
While it’s unlikely that we will see many ultra-HNWIs, we do have good numbers of higher-income respondents. To ensure a balanced mix of high, mid and low income respondents, our panel partners offer a variety of incentives that are designed to appeal to respondents of all backgrounds, including wealthier segments who might be most motivated by the chance to help good causes through charity donations.
Racial identity, ethnicity and nationality
Collecting data on the racial and ethnic identity of the population isn’t legal in all countries, so we can't ask about it everywhere.
Even in markets where asking about race and ethnicity is legal, information on the size of and internet penetration among each group is hard to come by and can often be unreliable, making the setting of accurate quotas impossible. In such markets, relying on natural fallout is the best approach.
At present, we ask about racial identity, ethnicity or nationality in Australia, Canada, Hong Kong, Indonesia, Ireland, Israel, Malaysia, Saudi Arabia, Singapore, South Africa, Thailand, UAE, UK, USA and Vietnam. Additionally:
- In the UAE, we set quotas on nationality to ensure a balance between Emiratis and other nationals.
- In the USA, we set quotas and apply weights on both race and ethnicity.
- In Saudi Arabia, we’ve developed a specific weighting framework in which we apply interlocking weights on age, gender and nationality quotas to ensure accurate representation of Saudi Nationals, Arab Expats and Non-Arab Expats (for more information please contact your account manager).
On top of this, in more linguistically diverse markets, we field our surveys in multiple languages. This makes our surveys more accessible to respondents who may not speak the dominant language in a given country, thereby improving our ability to include people from a diverse range of backgrounds in our sample. More information on the languages available in each market can be found here.