How to Survey South India Consumers: Five States, Many Cities

The Census of India 2011 is the last full count. Telangana was formed on 2 June 2014 (Government of Telangana), so the 2011 tables list its districts, including Hyderabad, under Andhra Pradesh. The table treats the two together. We computed shares from the Census mother-tongue tables (C-16) for each state.

Read three things from it. First, Kerala has seven urban agglomerations of 10 lakh or more in the provisional 2011 table, against one in Karnataka. Second, Bangalore's urban agglomeration (84,99,399, provisional) is about 36% of Karnataka's 2011 urban population. Chennai's (86,96,010) is about 25% of Tamil Nadu's. A Karnataka sample drawn only from Bangalore speaks for Bangalore, not the state. Third, Karnataka's largest mother tongue is 66.5% of the state, and Urdu is 10.8%. In Andhra Pradesh as counted in 2011, Urdu is 8.9%.

These are counts of mother tongue and population, not of survey members. For city detail, see our guides to Bangalore, Chennai and Hyderabad.

Our targeting fields are city and tier, gender (male or female), age band and NCCS A1 to C1, plus behavioural segments. Region is not a single field. You reach South India by naming the southern cities you want, and a study can be targeted at any Indian city based on your requirement.

Bangalore, Chennai and Hyderabad are the big southern cities, but they are examples, not the limit. Coimbatore, Kochi, Visakhapatnam, Mysuru, Madurai or any other town can be named too. How many members live in a given city decides whether it can fill your quota, so name each city and ask for a feasibility count before you plan the sample. Offline households are not reached, since the panel is app-based.

Hyderabad is in Telangana. Bangalore is in Karnataka. Chennai is in Tamil Nadu. Those three cover three states. For Kerala and Andhra Pradesh, name cities such as Kochi and Visakhapatnam. Do not promise a five-state result until you have a feasibility count for each city.

Interlocking quotas work across gender, age and city. State is not a quota field, so map each state to its cities and set quotas on the cities. Interlock city and gender. Set age as a simple quota, because city by gender by age cells get small quickly. See survey audience targeting.

Illustrative example. Suppose you need 600 completes and want to read each of six cities on its own, spread across the five states. Each cell is subject to a feasibility count.

Margin: 1.96 × √(0.25 ÷ n), which is ±9.8 points at n = 100 and ±4.0 at n = 600. It assumes a simple random sample and a 50% share, so treat it as a rough guide. At n = 50 per gender within a city it is ±13.9 points, too wide for a precise city-by-gender read.

If you also want a pooled figure for the three largest cities only, weight to the 2011 provisional agglomeration populations: Chennai 86,96,010, Bangalore 84,99,399 and Hyderabad 77,49,334, a total of 2,49,44,743. The shares are 34.9%, 34.1% and 31.1%. Against an equal third each, the weights are about 1.05, 1.02 and 0.93. Population is an old proxy for the member base, so state the choice in your report.

Surveys run in as many Indian languages as the study needs. Confirm that the language each of your quota cells needs is covered before you commit. Census mother tongue is a guide to who speaks what at home, but your questionnaire needs the language people read comfortably.

Rules that matter more in this region than in a one-language market: translate the screener as carefully as the questions, because a mistranslated option changes who qualifies. Keep brand and model names in the form shoppers use. Let respondents choose their language at the start. Check a few completes from each language by eye. Our multilingual survey guide goes deeper.

The brief: "People in South India who are interested in cars." "Interested" is vague, so define it. This screener qualifies people who plan a purchase, and tags owners for a second read.

1. Which of these does your household own? (Car, two-wheeler, bicycle, none of these). Tag car owners.
2. Which of these might you buy in the next 12 months? (Car, two-wheeler, phone, television, none of these). Continue if car.
3. When might you buy? (Within 3 months, 3 to 6 months, 6 to 12 months, later than 12 months). Qualify if within 12 months.
4. New or used? (New, used, not sure).
5. Which fuel types would you consider? (Petrol, diesel, CNG, electric, hybrid, not sure).
6. Which cars have you looked at? (A list of real models, one decoy model, none).
7. Which city do you live in? City is known from the panel profile, so ask only to verify it against the quota cell.

Rule: qualify on Q2 car and Q3 within 12 months, remove anyone who ticks the decoy in Q6, and route by Q1 into an owner or non-owner group. A looser rule, any interest in cars, gives more people but a vaguer read. The automobile research guide covers purchase stages and fuel choice.

Illustrative incidence. If 15% of people screened qualify, 600 completes need 600 ÷ 0.15 = 4,000 people screened. A city cell of 100 needs 100 ÷ 0.15 = about 667 in that city. If a city fills slowly, merge it with a nearby city or name another one.

1. For each city I name in the five states, what is the feasible count?
2. Is region a single field, or do I name cities?
3. Which languages can the survey run in, city by city?
4. Can I interlock city and gender, and what is the smallest cell?
5. How is each respondent verified? See verified respondents.
6. What incidence do you assume for my screener?

See also the audience hub and Tier 2 and Tier 3 research.

The panel is smartphone-only and India only. We publish no counts for any state, city or interest, and the Census figures above describe the population, not our members. Rural and offline households need face-to-face fieldwork. A five-state read needs at least one named city per state, each with a feasible count.

If your job is an online consumer survey in southern cities, this is the 600-response study above at list price. Screened-out respondents do not count towards a response bundle.

Pro is sold in bundles of 400 to 1,000 responses and costs 20% less per response than Plus. Enterprise pricing is custom.

The SuperJ network has a 25M+ consumer footprint. The survey-eligible panel is a subset, and availability depends on your targeting. Verification details are on /panel-methodology and pricing.

The limit: the app reaches smartphone users only.