Key points: primary and secondary data; scales of measurement; sampling (probability and non-probability); tabulation, classification, frequency distribution; graphs and diagrams (line, bar, pie, histogram, ogive, climograph, hythergraph, wind rose, Lorenz curve, scatter, triangular); central tendency (mean, median, mode, GM, HM); dispersion (range, QD, MD, SD, CV); normal curve, skewness, kurtosis; correlation and regression; tests (chi-square, t); spatial statistics (nearest neighbour, Gini, location quotient, PCA).
1. Data ⭐⭐
Primary data | Secondary data |
Collected first-hand by the investigator for a purpose | Already collected and published by others |
Methods: personal observation, interview, questionnaire (mailed, self-filled), schedule (filled by enumerator), field survey and measurement, case study, participatory methods, remote sensing / GPS | Sources: Census of India, NSSO/NSO surveys (PLFS, NFHS), Statistical Abstracts, IMD, CWC, Agricultural Census, Livestock Census, Economic Survey, World Bank, UN, published reports, maps, satellite archives |
Original, specific, costly and slow | Cheap, quick, large coverage; may not fit exactly; check reliability |
Scales (levels) of measurement ⭐⭐⭐
Scale | Property | Example | Valid statistics |
Nominal | Categories only (names) | Soil type, religion, land use | Mode, frequency, chi-square |
Ordinal | Ranked order | Town size class, rank-size of settlements, preference | Median, percentiles, Spearman's rank |
Interval | Equal intervals, arbitrary zero | Temperature in °C/°F, altitude above MSL, dates | Mean, SD, Pearson's r |
Ratio | Equal intervals and a true zero | Kelvin temperature, rainfall, income, distance, population | All, including ratios |
2. Sampling ⭐⭐⭐
- Population (universe) = all units; sample = selected part used to infer about the population. Advantages: saves time and cost; necessary for destructive tests; can be more accurate in practice.
Probability sampling | Non-probability sampling |
Simple random – every unit has an equal chance (lottery, random numbers); best for homogeneous populations | Purposive / judgement – chosen deliberately; useful when little is known about the population |
Systematic – every k-th unit after a random start (k = N/n) | Convenience – easily available units |
Stratified random – divide into homogeneous strata (e.g., plains, hills; gender), then sample randomly from each (proportionately or equally); best for heterogeneous areas | Quota – fixed numbers from categories, chosen non-randomly |
Cluster – random selection of groups (villages, blocks), then all units in them | Snowball – respondents refer others (hidden populations) |
Multi-stage – district → block → village → household | ㅤ |
- Spatial sampling: point, line (transect), area (quadrat) samples – random, systematic or stratified.
- Sampling error decreases as sample size increases; standard error of mean = s / √n.
3. Processing: Classification and Tabulation
- Classification: grouping data by common characteristics – qualitative (attributes), quantitative (class intervals), temporal (time series), spatial (geographical).
- Frequency distribution: raw data → classes with frequencies (counted with tally marks).
- Class interval types: exclusive (10–20, 20–30 – upper limit excluded) and inclusive (10–19, 20–29).
- Number of classes – Sturges' rule: k = 1 + 3.322 log N.
- Mid-value = (lower + upper limit) ÷ 2; cumulative frequency (less than / more than); relative frequency = f/N × 100.
- Tabulation: arranging data in rows and columns – parts: table number, title, captions (column headings), stubs (row headings), body, head-note, footnote, source. Simple (one-way), double, complex tables.
- For mapping, class intervals may be equal, quantile, natural breaks, standard deviation or arithmetic/geometric progression; variable (unequal) intervals suit highly skewed data.
4. Graphs and Diagrams ⭐⭐⭐
Type | Best for |
Line graph | Change over time – population growth, decadal growth, temperature trend |
Polygraph | Several lines on one graph (compare growth of cities) |
Bar diagram (simple, multiple, compound/divided) | Comparison of quantities – production by states, rainfall by month |
Pie / sector diagram (divided circle) | Parts of a whole – land use shares, occupational structure; angle = value/total × 360° |
Histogram | Continuous (grouped) frequency distribution – adjacent bars, area ∝ frequency |
Frequency polygon / curve | Joining mid-points of histogram tops |
Ogive (less than / more than) | Cumulative frequency – read median and quartiles |
Climograph (Griffith Taylor) | Wet-bulb temperature against relative humidity – human comfort |
Hythergraph | Temperature against rainfall for 12 months |
Ergograph | Seasonal human activity (crop calendar) with climate |
Wind rose / star diagram | Wind direction and frequency |
Scatter diagram | Relationship between two variables; natural groupings |
Triangular (ternary) graph | Three components adding to 100% – soil texture, occupational structure (primary–secondary–tertiary) |
Lorenz curve | Inequality / concentration – cumulative % of two variables vs the 45° line of equality; Gini coefficient = area between curve and line ÷ area under the line (0 = equality, 1 = total inequality) |
Band graph, semi-log graph (rates of change), rank-size graph (log–log), age–sex pyramid, isopleth graph | Various |
One-dimensional (bars), two-dimensional / area diagrams (squares, rectangles, circles, pies), three-dimensional / volume diagrams (cubes, spheres, blocks) | Proportional representation; spheres for very large ranges |
5. Measures of Central Tendency ⭐⭐⭐
Measure | Formula / method | Features |
Arithmetic mean | x̄ = Σx/N; grouped: Σfx/Σf (using mid-points); assumed-mean (short-cut) and step-deviation methods | Uses all values; affected by extreme values; Σ(x − x̄) = 0; Σ(x − x̄)² is minimum; basis of SD, regression; weighted mean for centres of population |
Median | Middle value of ordered data; grouped: L + [(N/2 − cf)/f] × h | Positional average; not affected by extreme values; read from ogive |
Mode | Most frequent value; grouped: L + [(f1 − f0)/(2f1 − f0 − f2)] × h | Most typical value; unaffected by extremes; empirical: Mode = 3 Median − 2 Mean |
Geometric mean | nth root of the product | Growth rates, ratios, index numbers |
Harmonic mean | N / Σ(1/x) | Averaging rates such as speeds |
Quartiles, deciles, percentiles | Q1, Q3 = L + [(N/4 − cf)/f] × h etc. | Positional measures |
- AM ≥ GM ≥ HM. Spatial "averages": mean centre (weighted mean of coordinates – shift of population centre), median centre.
- Worked example: classes 10–20 (6), 20–30 (20), 30–40 (5) → mean = (15×6 + 25×20 + 35×5)/31 = 765/31 ≈ 24.7.
6. Measures of Dispersion ⭐⭐⭐
Measure | Formula | Notes |
Range | Max − min | Simplest; coefficient = (L − S)/(L + S) |
Quartile deviation | (Q3 − Q1)/2 | Semi-interquartile range |
Mean deviation | Σ | x − x̄ |
Variance | σ² = Σ(x − x̄)²/N | ㅤ |
Standard deviation (Karl Pearson, 1893) | σ = √[Σ(x − x̄)²/N] – positive square root of the mean of squared deviations from the mean | Most used; independent of change of origin, but affected by change of scale |
Coefficient of variation | CV = (σ / x̄) × 100 | Relative, unit-free – compares variability; rainfall variability; higher CV = less consistent |
- Absolute measures: range, QD, MD, SD; relative measures: coefficients (CV etc.).
Normal distribution
- Bell-shaped, symmetric; mean = median = mode. ±1σ = 68.27% (34.13% each side), ±2σ = 95.45%, ±3σ = 99.73% of values. Standard score z = (x − x̄)/σ.
- Skewness (asymmetry): positive (tail to right; mean > median > mode), negative (tail to left). Karl Pearson's coefficient = (mean − mode)/σ.
- Kurtosis (peakedness): leptokurtic (more peaked, β2 > 3), mesokurtic (normal, = 3), platykurtic (flat, < 3).
7. Correlation and Regression ⭐⭐⭐
- Correlation = degree and direction of association between two variables; shown first by a scatter diagram.
- Karl Pearson's product-moment coefficient: r = Σ(x − x̄)(y − ȳ) / √[Σ(x − x̄)² Σ(y − ȳ)²] = covariance / (σx σy).
- Range −1 to +1; +1 perfect positive, −1 perfect negative, 0 no linear relation.
- Independent of origin, scale and units; parametric (interval/ratio data, bivariate normal); sensitive to outliers; r² = coefficient of determination (share of variation explained); r = √R² (with sign).
- Spearman's rank correlation (non-parametric, ordinal data): ρ = 1 − 6ΣD² / [n(n² − 1)].
- Regression estimates (predicts) one variable from another: Y = a + bX; b (slope / regression coefficient) = Σxy/Σx² (deviations); the line passes through (x̄, ȳ) so a = ȳ − b x̄.
- Example: x̄ = 23, ȳ = 94.5, b = 1.5 → a = 94.5 − 34.5 = 60.
- Assumptions (OLS): linearity, normally distributed residuals, constant variance (homoscedasticity), independent residuals.
- Good fit → R² near 1. Residuals (observed − predicted) can be mapped to find anomalies.
- Multivariate: multiple regression, factor analysis / principal component analysis (reduce many correlated variables to a few components; coefficients are loadings; eigenvalues; used for regionalisation and development indices), cluster analysis (agglomerative or divisive grouping).
8. Tests and Spatial Statistics
Technique | Use |
Chi-square (χ²) test | Non-parametric; nominal/frequency data; goodness of fit (e.g., against a uniform model) or independence; H₀: no significant difference between observed and expected; reject H₀ if computed χ² > table value |
t-test | Small samples (n < 30) with unknown population SD – compare a sample mean with a population mean |
Z-test | Large samples / known population SD |
F-test / ANOVA | Compare variances / several means |
Nearest neighbour analysis (Clark and Evans, 1954; King, 1962 for settlements) | Rn = observed mean distance ÷ expected mean distance (expected = 0.5/√density). Rn = 0 clustered; 1.0 random; 2.15 perfectly uniform (triangular) |
Lorenz curve and Gini coefficient | Inequality / concentration of a distribution (e.g., ST population, income) |
Location quotient | (Local share) ÷ (national share) – regional specialisation; > 1 means concentration |
Index of dissimilarity / segregation | Unevenness of two groups |
Composite indices | Ranking (summation of ranks – simple, robust for skewed data), z-scores, PCA for development levels |
Time series | Secular trend (population growth), seasonal/periodic (milk production), cyclical (long-term climate change), irregular (earthquakes); moving averages |
Network indices | Beta = e/v; Gamma = e/3(v − 2); Pi = total length ÷ diameter; alpha (circuits) |
Gravity / spatial interaction models | Tij = k PiPj / dij^b (Sij = separation between origin and destination) |
⚡ Quick Revision
- Kelvin = ratio; °C, altitude = interval; town-size rank = ordinal; soil type = nominal.
- Stratified random → heterogeneous; simple random → homogeneous; snowball, convenience, purposive, quota = non-probability.
- Histogram → continuous frequency; ogive → cumulative; pie → parts of whole; line → growth over time; Lorenz → inequality.
- Median not affected by extremes; mean Σ(x − x̄) = 0; GM → growth rates; HM → speeds.
- SD = √variance; CV = SD/mean × 100 (relative); ±1σ = 68.27%.
- r from −1 to +1, independent of origin and scale; Spearman for ranks; a = ȳ − b x̄.
- Chi-square non-parametric; Rn: 0 cluster, 1 random, 2.15 uniform.
📝 Practice MCQs
- The measure most suitable to compare rainfall variability of two stations is: (a) SD (b) range (c) coefficient of variation (d) mean deviation
Answer
(c) Coefficient of variation
- Which average is not affected by extreme values? (a) Mean (b) Median (c) Geometric mean (d) Harmonic mean
Answer
(b) Median (mode too)
- An Rn value of 2.15 indicates: (a) clustered (b) random (c) perfectly uniform (d) linear pattern
Answer
(c) Perfectly uniform
- If Y = a + bX, x̄ = 5.5, ȳ = 3.5, b = 1.5, then a = (a) 3.75 (b) –4.75 (c) 1.75 (d) –5.75
Answer
(b) –4.75
- Which sampling method is NOT a probability method? (a) Stratified (b) Cluster (c) Snowball (d) Systematic
Answer
(c) Snowball
- A curve more peaked than the normal curve is: (a) platykurtic (b) mesokurtic (c) leptokurtic (d) skewed
Answer
(c) Leptokurtic