Surveys, Demography and Forecasting: Classification, Sampling, Projection and Planning Analysis

Patrick Geddes’s rule, “survey before plan” (survey, analysis, plan), still organises the planning process. A survey collects facts that the plan then uses; the type of survey follows the question.

Choices that cut across the types. A census (complete enumeration) covers every unit; a sample survey covers a part and infers the whole, and is quicker and cheaper but carries sampling error. Primary data are collected fresh; secondary data come from the Census, other national surveys, municipal and revenue records. A cross-sectional survey is a single snapshot; a longitudinal or panel survey follows the same units over time. A reconnaissance or rapid survey sketches the picture; a detailed survey measures it.

Tools. A schedule is filled in by the enumerator who asks the questions; a questionnaire is filled in by the respondent. Questions may be structured (fixed choices) or open. Interviews can be individual or with key informants (ward members, shopkeepers, officials); a focus group discussion gathers a small group to explore reasons and attitudes in depth. Observation may be structured, with counts and checklists. Participatory rural or learning appraisal (PRA and PLA) uses transect walks, social and resource mapping, ranking, seasonal calendars and Venn diagrams so that communities build their own picture. Field instruments include total stations, GPS, drones and tablets running survey apps.

Quality of a survey. The cycle is: objective, unit and frame, instrument, pilot test, fieldwork with supervision, data checking, analysis, report. Errors are of two kinds: sampling error, which falls as the sample grows, and non-sampling error (measurement, non-response, coverage, recording), which does not fall with sample size and can be large even in a census. An instrument is reliable if it gives the same result on repetition and valid if it measures what it claims to. Answers are measured on nominal, ordinal, interval or ratio scales, and a Likert scale (agree to disagree) is ordinal.

A sampling frame is the list from which units are drawn (a ward’s household list, a voters’ roll, a plot register). Probability sampling gives every unit a known, non-zero chance, so the sampling error can be calculated; non-probability sampling does not, so the result cannot be generalised with a stated error.

Systematic and stratified arithmetic. From a list of N = 1,200 households choose n = 60: the interval is k = 1,200 / 60 = 20; if the random start in 1 to 20 is 7, the sample is 7, 27, 47 and so on, and the tenth unit is 7 + 9 × 20 = 187. For stratified proportional allocation the stratum sample is n_h = n × N_h / N: with N = 5,000 households in strata of 2,000, 2,000 and 1,000 and n = 200, the allocation is 80, 80 and 40. Optimum (Neyman) allocation gives more to strata that are large or more variable, n_h proportional to N_h × S_h: with the same strata and standard deviations of 10, 20 and 40, the products are 20,000, 40,000 and 40,000, so n = 200 splits as 40, 80 and 80.

Sample size. To estimate a proportion within a margin of error e at a confidence level with normal deviate z, the first estimate is n_0 = z² p (1 − p) / e². With p = 0.5, which gives the largest and so the safest n, z = 1.96 (95 percent) and e = 0.05: n_0 = 1.96² × 0.25 / 0.05² = 384.16, which is rounded up to 385. If the population is small, a finite population correction shrinks it: n = n_0 / (1 + n_0 / N); for N = 2,000, n = 384.16 / 1.19208 = 322.3, so 323. The simplified Yamane formula n = N / (1 + N e²) is also used and gives 334 here, so a question must say which formula it wants. To estimate a mean, n = (z σ / E)²: with σ = 12, E = 2 and z = 1.96, n = (11.76)² = 138.3, so 139.

Demography is the statistical study of human populations: their size, distribution and composition, and the three processes that change them, fertility, mortality and migration. For planning it supplies the base population and its structure, which fix the demand for housing, schools, health care, water and jobs. The balancing equation is P_t = P_0 + B − D + I − O: births, deaths, in-migrants and out-migrants.

Worked examples. A town of 240,000 people (mid-year) records in a year 6,000 births, 2,400 deaths, 1,200 in-migrants and 600 out-migrants. CBR = 6,000 / 240,000 × 1,000 = 25; CDR = 10; natural increase = 15 per 1,000 = 1.5 percent; net migration = +600 = 2.5 per 1,000; the overall growth rate is (6,000 − 2,400 + 1,200 − 600) / 240,000 = 1.75 percent a year. For fertility, take age-specific rates per 1,000 women of 30, 140, 150, 90, 50, 15 and 5 for ages 15–19 to 45–49: their sum is 480, so TFR = 5 × 480 / 1,000 = 2.4 children per woman. For a population of 100,000 with 30,000 aged 0–14, 62,500 aged 15–64 and 7,500 aged 65 and over, the dependency ratio is (30,000 + 7,500) / 62,500 × 100 = 60, made up of a young part of 48 and an old part of 12. A city growing at r percent a year doubles in about 70 / r years.

India’s conventions. The Census of India gives the sex ratio as females per 1,000 males (943 in Census 2011, and 919 for children aged 0–6), whereas the international convention is males per 100 females. A main worker has worked at least 183 days in the preceding year; a marginal worker, less. The work participation rate is workers ÷ population × 100. India’s total fertility rate has fallen to about 2.0 in the fifth National Family Health Survey, just below replacement level. The shape of the population pyramid tells the story: a wide base is a young, expanding population; a narrow base with a bulge in the middle is ageing and brings a demographic dividend while the working-age share is high; the demographic transition runs from high birth and death rates, through falling deaths and rapid growth, to low birth and death rates.

A projection extends past population into the future under stated assumptions, to size the housing, schools, water and roads for the plan period. Every method below is tried on one data set: a town’s census populations of 100,000 (1991), 120,000 (2001) and 144,000 (2011), so the first increase is 20,000, the second 24,000, and the growth is 20 percent in each decade.

Here n counts decades from the last census (n = 1 is 2021, n = 2 is 2031); in the incremental formula the term n(n + 1)/2 equals 1 for n = 1 and 3 for n = 2, so for 2031 it is 144,000 + 2 × 22,000 + 3 × 4,000 = 200,000. The exponential result equals the geometric one over a whole decade because e^0.1823 = 1.2; the two differ in the annual rate they imply, 1.823 percent continuous against 1.2^0.1 − 1 = 1.84 percent compound. Note that the three give 166,000, 170,000 and 172,800 for 2021: the arithmetic method assumes the growth will not accelerate, the geometric assumes the percentage never falls.

Other methods. The graphical method plots census points and extends the curve by eye. The graphical comparison method follows the curve of a larger, older, similar city that grew earlier. The logistic curve is S-shaped: growth first accelerates, then slows toward a saturation population P_s, P = P_s / (1 + e^(a + b t)); with three equally spaced censuses P_0, P_1, P_2, the saturation is P_s = [2 P_0 P_1 P_2 − P_1² (P_0 + P_2)] / (P_0 P_2 − P_1²). The ratio (or share) method takes the town’s share of a larger area, such as the district or state, whose projection already exists, extends the share, and multiplies: a town with 0.8 percent of the state in 2001 and 1.0 percent in 2011 reaches 1.2 percent in 2021 by a straight-line extension, and with the state at 25 million it has 0.012 × 25,000,000 = 300,000.

Cohort-component method. The population is split into age-sex cohorts and moved forward in steps of five or ten years. Each cohort is multiplied by its survival ratio from a life table and adjusted for net migration by age; births come from the age-specific fertility rates applied to the women in each child-bearing cohort and are then survived to form the new youngest cohort. Example: 12,000 females aged 15–19 with a five-year survival ratio of 0.995 and net in-migration of 200 become 12,000 × 0.995 + 200 = 12,140 aged 20–24 five years later. It is the only method that tells how many children, workers and elderly there will be, so it is used for school, health and labour-force planning.

An estimate is a best figure for the present or the past between two counts, for example a town’s population in 2006 from the censuses of 2001 and 2011; a forecast is a figure for the future. For 2006 from 120,000 (2001) and 144,000 (2011), straight-line interpolation gives 120,000 + 0.5 × 24,000 = 132,000, and geometric interpolation 120,000 × 1.2^0.5 = 131,453. Forecasting methods fall in three families: extrapolative (trend lines, curves, time series with trend, seasonal and cyclical parts, moving averages), causal (regression, economic base, input–output, spatial interaction and land use models) and judgemental (expert panels, scenario writing and the Delphi method, in which experts answer anonymously over rounds and see the group’s summary each time until opinion converges).

Economic base and the base multiplier. Employment is split into basic jobs, which sell outside the area and bring in income, and non-basic (service) jobs, which serve the local population. Growth is led by the basic sector, and the base multiplier = total employment ÷ basic employment. A town with 4,000 basic and 6,000 non-basic jobs has total 10,000, so the multiplier is 2.5; if the ratio holds, 800 new basic jobs raise total employment by 800 × 2.5 = 2,000. The weakness is the assumption that the ratio and the basic–non-basic split stay fixed.

Location quotient. LQ = (local employment in industry i ÷ local total employment) ÷ (national employment in i ÷ national total). LQ above 1 marks a specialisation and a likely export (basic) industry; below 1 suggests a net importer; LQ = 1 is local self-sufficiency. The excess employment over the national share, e_i × (1 − 1/LQ), estimates the basic jobs. Example: a city with 2,000 of 20,000 jobs in an industry (10 percent) against a national share of 1.2 million of 20 million (6 percent): LQ = 10 / 6 = 1.67 and basic jobs = 2,000 × (1 − 1/1.667) = 800.

Shift-share analysis. The change in a region’s employment in an industry over a period is split into three parts: the national growth effect (what it would have gained by growing at the national all-industry rate), the industrial mix effect (the national growth of that industry above or below the national average) and the regional shift (competitive) effect (how much better or worse the region did than the nation in that industry). They add up to the actual change. Example: textile jobs rise from 2,000 to 2,500 (+25 percent); the nation’s all-industry growth is 8 percent and its textile growth 15 percent. National growth = 2,000 × 0.08 = 160; industrial mix = 2,000 × (0.15 − 0.08) = 140; regional shift = 2,000 × (0.25 − 0.15) = 200; the total is 500, as observed.

Input–output analysis (Leontief), in principle. A table records how much each sector sells to every other sector and to final demand. The technical coefficient a_ij is the input from sector i needed per unit of output of sector j. Gross output x satisfies x = A x + f, so x = (I − A)⁻¹ f, where the Leontief inverse (I − A)⁻¹ gives the total, direct plus indirect, output required per unit of final demand. For two sectors with a₁₁ = 0.2, a₁₂ = 0.3, a₂₁ = 0.4, a₂₂ = 0.1 and final demands 120 and 210, the outputs are 285 and 360. Column sums of the inverse give output multipliers; adding households as a sector gives the larger induced (Type II) effect.

Goal formulation turns the findings of the surveys into direction. The ladder runs from vision (an image of the desired future), to goals (broad, qualitative ends, such as “a healthy, inclusive city”), objectives (specific, measurable and time-bound steps towards a goal), targets (the quantified level, such as “universal piped water by 2035”), and then policies, strategies and projects as the means. Goals come from problem and needs analysis, SWOT, stakeholder consultation, Delphi rounds and public meetings; conflicts among them (growth against environment, for example) are settled by trade-off, weighting or the goals-achievement matrix.

Scenario analysis answers the question “what might happen, and how would the plan fare?” A scenario is not a prediction but an internally consistent story of a plausible future built on stated assumptions about key drivers and uncertainties (migration, economic growth, land policy, fuel prices, climate). Planners usually set a business-as-usual trend scenario against alternatives such as compact growth, dispersed growth and corridor or polycentric growth, and then test each plan option, with a sensitivity analysis, against all of them. Scenarios may be exploratory (where do present trends lead?) or normative (what must happen to reach a desired future, worked backwards). A low, medium and high population projection is the simplest scenario set.

Planning indicators are measurable proxies for a condition: they are tracked as inputs (money, staff), outputs (taps connected), outcomes (hours of supply) and impacts (waterborne disease). A good indicator is specific, measurable, available, relevant and comparable over time; examples are density, open space per person, litres per capita per day of water, the share of waste processed, and composite indices such as the Human Development Index. In India the Ministry’s Service Level Benchmarks (2008) set indicators for water supply, sewerage, solid waste and storm drainage, and the Sustainable Development Goal 11 framework gives global urban indicators.

Norms and standards. A norm is an accepted value of provision or performance, usually per capita or per unit of population, drawn from experience; a standard is the level a plan commits to or an authority prescribes (the national guidelines URDPFI and the National Building Code are the Indian sources). Types are space standards (area per thousand people), service-level standards (litres per head per day), locational or catchment standards (distance, time, population served), performance standards (noise, emissions) and design standards. The arithmetic is simple: area required = projected population × per-capita standard. For an assumed standard of 12 sq m of open space per person and a projected 300,000 people, the requirement is 3,600,000 sq m = 360 hectares. Standards fit unlike contexts badly if applied rigidly: a norm set for a large new town is not automatically right for a dense old city, so values are checked against local conditions.

Plans are lived in, so planners study how people perceive and use the city. Perception is how people sense and interpret surroundings; cognition is how they organise that knowledge in the mind, as a cognitive (mental) map; behaviour is what they do. Kevin Lynch, in “The Image of the City” (1960), asked residents of Boston, Jersey City and Los Angeles to sketch their city and describe routes, and found that people build the image from five elements. A city is legible or imageable when these elements are clear and connected.

Related ideas shape design. Edward Hall’s proxemics treats personal space as nested zones (intimate, personal, social, public). Territoriality and Oscar Newman’s defensible space link clear ownership and visibility to safety. William Whyte’s studies of urban plazas found that people use spaces with sittable edges, sun, water, food and the sight of other people. Place attachment is the emotional bond to a locale, which displacement breaks. Space-syntax analysis relates how connected a street is to how much it is used.

Methods. Perception is measured by opinion and satisfaction surveys, semantic differential and Likert scales, sketch-map and photo-based exercises, interviews and focus groups, and citizen report cards that rate services. Behaviour is measured by observation: behaviour mapping, pedestrian and activity counts, and travel diaries. The two must be read together because of the gap between stated preference (what people say they would do) and revealed preference (what they are seen to do). People often say that they would use buses or walk if conditions improved, yet observed counts differ; the observed behaviour is the better guide for demand, the stated for attitudes and unmet needs.