Social work research methods and statistics

In short

Social work research is the systematic study of social problems, client needs, intervention processes and programme outcomes, using the same scientific logic as other social sciences but tied to practice and to social justice. It covers the whole chain — problem formulation, design, sampling, tools, fieldwork, coding, statistical analysis, interpretation and report writing — plus the ethics of working with vulnerable people. In Indian MSW programmes it is a core paper in every university (TISS, DU, Jamia, Christ, Loyola, Madras, IGNOU), is the base of the compulsory dissertation/research project, and is tested in UGC-NET Social Work, M.Phil/PhD entrances and Ph.D coursework. Employers test it in interviews for monitoring-and-evaluation officers, research associates, programme officers and CSR roles at NGOs, foundations, UNICEF/UNDP projects, ICSSR and NIMHANS-type institutions, where proposal writing, baseline surveys and data analysis in Excel/SPSS are daily work.

Explained

What social work research is, and its types

• Definition: application of scientific method to problems and questions arising in social work practice — to build knowledge, test interventions and inform policy. It is distinguished by its purpose (improving welfare), not by a unique method.
• Social research vs social work research: social research studies society broadly; social work research is practice-focused — it studies clients, workers, agencies, interventions and outcomes.
• By purpose: *basic/pure* (theory building), *applied* (solve a field problem), *action research* (research and intervention together, with participants), *evaluative* (judges a programme — treated in the companion topic).
• By objective: *exploratory/formulative* (little known; flexible, builds hypotheses), *descriptive* (accurately describes who/what/how many; surveys), *diagnostic* (causes and treatment of a problem), *explanatory/causal* (tests cause–effect), *experimental*.
• By data type: *quantitative* (numbers, deductive, tests hypotheses), *qualitative* (meanings, inductive, builds theory), *mixed methods* (triangulation; sequential or concurrent designs).
• By time: *cross-sectional* (one point in time), *longitudinal* — panel (same people), cohort (same age/event group), trend (same population, different samples); *retrospective* (looks back).
• Scientific method: empirical, systematic, objective, replicable, self-correcting, aims at generalisation; its limits in social work are human reflexivity, unobservable variables and ethical restrictions on experimentation.

Steps of the research process

1. Select and formulate the problem — from practice, literature or policy; narrow a vague interest into a researchable question.
2. Review literature — journals (*Indian Journal of Social Work* from TISS, *Social Change*, *Social Work Education*), Shodhganga (INFLIBNET) theses, government reports (NFHS, NSS/PLFS, Census); identifies gaps and tools.
3. Define concepts and variables — conceptual definition, then operational definition (how it will be measured); independent, dependent, intervening, extraneous/confounding, moderator variables.
4. Frame objectives and hypotheses — general objective plus specific measurable objectives; hypotheses only where the state of knowledge allows.
5. Choose the design — exploratory, descriptive, experimental, single-system, case study, ethnographic.
6. Decide universe, sampling frame, unit and sample size.
7. Construct tools and pre-test/pilot — check wording, sequence, time taken, reliability.
8. Collect data — fieldwork, rapport, supervision, field notes.
9. Process — editing, coding, a code book, data entry, cleaning, tabulation.
10. Analyse and interpret — descriptive statistics, then inferential tests; link back to objectives.
11. Report — chapters: introduction, review of literature, methodology, analysis and interpretation, findings and suggestions; references in APA style, appendices with the tool.
12. Research proposal/synopsis: title, background, statement of the problem, significance, objectives, hypotheses, methodology, limitations, chapter scheme, time frame (Gantt), budget, references — the format ICSSR, NGOs and university research committees expect.

Hypotheses, variables and measurement

• Hypothesis: a tentative, testable statement of relationship between two or more variables. Must be clear, specific, empirically testable and related to theory.
• Types: *research/working*, *null* (H₀: no difference/relationship — the one actually tested), *alternative* (H₁), *directional* (one-tailed) vs *non-directional* (two-tailed).
• Errors: Type I (α) — rejecting a true H₀ (false positive); Type II (β) — accepting a false H₀. Power = 1 − β. Common α in social work: 0.05 and 0.01.
• Levels of measurement: nominal (religion, caste category, sex), ordinal (ranks, Likert positions), interval (standardised scores, temperature), ratio (income in ₹, age in years, number of children) — the level decides which statistic is permissible.
• Scales: Likert summated rating, Thurstone equal-appearing intervals, Guttman cumulative/scalogram, Bogardus social distance, semantic differential, sociometry (sociograms for group work).
• Reliability (consistency): test–retest, split-half, parallel forms, inter-rater; Cronbach's alpha for internal consistency — above 0.70 generally acceptable.
• Validity (measuring the right thing): face, content, criterion (concurrent and predictive), construct (convergent and discriminant). A tool can be reliable yet invalid; it cannot be valid without being reliable.
• Qualitative equivalents (Lincoln and Guba): credibility, transferability, dependability, confirmability — achieved by triangulation, member checking, audit trail, prolonged engagement, reflexivity.

Sampling

• Terms: *universe/population*, *sampling frame* (e.g. the ICDS beneficiary register, an SHG list), *sampling unit*, *element*, *parameter* (population value) vs *statistic* (sample value), *sampling error*, *non-sampling error* (coverage, non-response, measurement).
• Probability methods: simple random (lottery, random number table), systematic (every kᵗʰ, k = N/n), stratified (strata by caste, sex, block; proportionate or disproportionate — better precision when strata differ), cluster (sample whole villages/wards; cheap but larger error), multi-stage (district → block → village → household, as in NFHS), PPS (probability proportional to size).
• Non-probability methods: purposive/judgemental, quota, convenience/accidental, snowball (hidden populations — sex workers, drug users, undocumented migrants), theoretical sampling (grounded theory). Cannot support statistical generalisation but are often the only feasible route.
• Sample size depends on variability in the population, precision wanted, confidence level, design (design effect for clusters), expected non-response and budget — not on a fixed percentage of the universe.
• Saturation is the qualitative stopping rule: sampling continues until no new themes emerge; typical focus group is 6–12 participants, 2–3 groups per stratum.

Tools and methods of data collection

• Primary sources: observation, interview, questionnaire, schedule, FGD, case study, scales.
• Observation: participant vs non-participant, structured vs unstructured, controlled vs natural, overt vs covert (ethically problematic). Strength: behaviour as it happens; limits: observer bias, Hawthorne effect, small coverage.
• Interview: structured, semi-structured, unstructured/in-depth, key informant, focus group, narrative/life history. An interview guide lists themes; probing draws depth. Strengths: depth, rapport, suits low-literacy respondents; limits: interviewer bias, cost, time.
• Questionnaire vs schedule: a questionnaire is filled by the respondent (mail, online, Google Forms); a schedule is filled by the investigator while asking — standard in Indian field surveys where literacy varies.
• Question types: closed-ended (dichotomous, multiple choice, Likert, ranking), open-ended, contingency/filter questions. Avoid double-barrelled, leading and loaded wording.
• Secondary sources: Census of India, NFHS-5 (2019-21), NSSO/PLFS, SRS, HMIS, UDISE+, NCRB *Crime in India*, district statistical handbooks, agency case records, NITI Aayog SDG India Index. Check definitions, reference period, base year and comparability; CPI base is 2024=100 from January 2026 and the GDP/IIP/WPI bases are 2022-23, so older series must be re-based before comparison.
• Qualitative approaches: case study (single/multiple, holistic), ethnography, phenomenology, grounded theory (open–axial–selective coding, constant comparison), narrative analysis, content analysis (manifest and latent), thematic analysis.
• Participatory methods (PRA/PLA): social and resource mapping, transect walk, seasonal calendar, daily activity chart, Venn/chapati diagram, wealth and well-being ranking, pair-wise ranking, timeline, problem tree, focus groups — widely used in rural projects, NRLM/SHG work and needs assessments.
• Single-system (single-subject) design — the practitioner-researcher design: repeated measurement of one client/group across phases. AB (baseline then intervention), ABA/ABAB (withdrawal/reversal), multiple baseline, changing criterion. Analysed visually (level, trend, variability) rather than by large-sample statistics.

Statistics for social work

• Descriptive statistics: frequency distribution, class intervals, percentage, cumulative percentage; graphs — bar, pie, histogram, frequency polygon, ogive, box plot.
• Central tendency: mean (interval/ratio; distorted by outliers such as one very high income), median (ordinal and skewed data — preferred for income), mode (nominal data).
• Dispersion: range, quartile deviation, mean deviation, variance, standard deviation, coefficient of variation (compares variability across different units).
• Shape: skewness (positive — income distributions), kurtosis; the normal curve — symmetrical, mean = median = mode, about 68 / 95 / 99.7 per cent of cases within 1 / 2 / 3 SD.
• Relative position: percentiles, quartiles, z-scores.
• Inferential logic: sampling distribution, standard error, confidence interval, significance level, degrees of freedom, p-value, one- vs two-tailed test. Significant does not mean important — report effect size too.
• Parametric tests (assume normality, interval data, homogeneity of variance): one-sample, independent and paired t-test; ANOVA (one-way, two-way) with post-hoc tests; Pearson's r; regression (simple and multiple).
• Non-parametric tests (nominal/ordinal or non-normal data): chi-square for goodness of fit and independence, Mann-Whitney U, Wilcoxon signed-rank, Kruskal-Wallis, Friedman, McNemar, Spearman's rho, Cramér's V, Fisher's exact test for small cells.
• Choosing a test: by the measurement level, number of groups, independent vs related samples and whether assumptions hold — this is the single most examined piece of judgement.
• Software: SPSS, R, PSPP (free), Excel, Epi Info, KoBoToolbox/ODK and CommCare for mobile data collection, NVivo/ATLAS.ti (and manual coding) for qualitative data.

Ethics, and the Indian institutional setting

• Core principles: voluntary participation, informed consent (written, in the respondent's language; assent plus guardian consent for children), no harm, confidentiality and anonymity, privacy, right to withdraw, no deception, debriefing, honest reporting, no plagiarism, data security, sharing findings with the community.
• Vulnerable groups need extra safeguards — children, persons with disability, persons with mental illness, survivors of violence, prisoners, tribal communities. Capacity and supported decision-making follow the Rights of Persons with Disabilities Act, 2016 and the Mental Healthcare Act, 2017.
• Mandatory reporting overrides confidentiality where a child is sexually abused (POCSO Act, 2012 — reporting is a legal duty), and research involving children must respect the Juvenile Justice (Care and Protection of Children) Act, 2015, which bars disclosure of a child's identity.
• Ethics review: Institutional Ethics Committee/IRB clearance; the ICMR National Ethical Guidelines for Biomedical and Health Research Involving Human Participants (2017) is the standard reference for health-related social research in India, with ICSSR guidance for social science research; NAAC and university research boards require plagiarism checks under the UGC (Promotion of Academic Integrity) Regulations, 2018.
• Data protection: the Digital Personal Data Protection Act, 2023 and the DPDP Rules, 2025 govern digital personal data; the Data Fiduciary duties (notice, consent, breach intimation, erasure) and the omission of IT Act s.43A take effect only from 13 May 2027, so until then IT Act s.43A and the SPDI Rules, 2011 apply to sensitive personal data held in digital form.
• Funding and dissemination: ICSSR (doctoral fellowships, research programmes), UGC, ICMR, state social welfare boards, CSR funds under the Companies Act, 2013 (s.135 and Schedule VII; the CSR Policy Rules, 2014 carry the impact-assessment requirement for large projects), bilateral and UN agencies. Research is used in CSR impact assessment, social audits (notably the VB-G RAM G Act, which replaced MGNREGA from 1 July 2026 with a 125-day entitlement) and RTI-based studies.

Key points

• Social work research — scientific study of needs, practice and programmes, aimed at improving welfare, not only at theory.
• Exploratory, descriptive, diagnostic, experimental — the four classical designs; the question decides the design, never the reverse.
• Operational definition — the statement of how an abstract concept (empowerment, stigma) will actually be measured.
• Hypothesis — testable statement of relationship; the null hypothesis is what a statistical test actually examines.
• Type I error — rejecting a true null (α); Type II — failing to reject a false null (β); power = 1 − β.
• Measurement levels — nominal, ordinal, interval, ratio; they determine the permissible statistic.
• Reliability — consistency across time, items or raters; validity — whether the tool measures the intended construct.
• Probability sampling permits generalisation and error estimation; non-probability does not, but suits hidden and hard-to-reach groups.
• Snowball sampling — referral chains for hidden populations; saturation — qualitative stopping rule.
• Schedule vs questionnaire — investigator-administered versus self-administered; the schedule copes with low literacy.
• Triangulation — using multiple methods, sources, investigators or theories to strengthen findings.
• Single-system design (AB, ABAB, multiple baseline) — the practitioner's own evaluation of one case with repeated measures.
• Mean, median, mode — mean for symmetric interval data, median for skewed data such as income, mode for nominal data.
• Standard deviation and coefficient of variation — absolute and relative measures of spread; CV compares unlike units.
• Chi-square — association between two categorical variables; needs expected frequency ≥ 5 in most cells.
• Pearson's r (interval, linear) vs Spearman's rho (ranks); correlation is not causation.
• Informed consent, confidentiality, no harm, IEC clearance — non-negotiable; POCSO reporting overrides confidentiality.
• Secondary data — Census, NFHS-5, PLFS, NCRB, UDISE+; always check definition, reference period and base year (CPI 2024=100 from January 2026).

Formulas and examples

Example 1 — mean, median, SD and CV of monthly incomes. Eight SHG members earn (₹/month): 4,000; 5,000; 6,000; 6,000; 7,000; 8,000; 9,000; 35,000.

• Σx = 4,000+5,000+6,000+6,000+7,000+8,000+9,000+35,000 = ₹80,000 → x̄ = 80,000/8 = ₹10,000.
• Median = average of 4th and 5th values = (6,000 + 7,000)/2 = ₹6,500. Mode = ₹6,000.
• Deviations from 10,000: −6,000; −5,000; −4,000; −4,000; −3,000; −2,000; −1,000; +25,000. Squares: 36 + 25 + 16 + 16 + 9 + 4 + 1 + 625 (all ×10⁶) = 732 × 10⁶.
• s = √(732×10⁶ / 7) = √(104.57×10⁶) ≈ ₹10,226. CV = 10,226/10,000 × 100 ≈ 102.3 per cent.
• Interpretation: one outlier (₹35,000) pulls the mean to ₹10,000 while five of eight members earn ₹7,000 or less — report the median for skewed income data, and the CV shows extreme inequality within the group.

Example 2 — chi-square test of independence (training and employment). Of 200 trainees from a livelihood project: among 120 who completed the full course, 84 got jobs; among 80 who dropped out, 36 got jobs.

• Expected: E₁₁ = 120×120/200 = 72; E₁₂ = 120×80/200 = 48; E₂₁ = 80×120/200 = 48; E₂₂ = 80×80/200 = 32.
• χ² = (84−72)²/72 + (36−48)²/48 + (36−48)²/48 + (44−32)²/32 = 144/72 + 144/48 + 144/48 + 144/32 = 2 + 3 + 3 + 4.5 = 12.5.
• df = (2−1)(2−1) = 1; critical value at α = 0.05 is 3.84 and at 0.01 is 6.63 → reject H₀; completion and employment are significantly associated. Effect size (phi) = √(12.5/200) ≈ 0.25, a modest association — significance alone does not prove the training caused the jobs, since dropping out may itself reflect prior disadvantage.

Example 3 — paired comparison of a self-esteem score (pre/post). Five adolescents score before and after a 10-week group work intervention:

• Σd = 6+3+6+3+7 = 25 → mean d̄ = 5.0 points.
• Deviations from 5: +1, −2, +1, −2, +2; squares 1+4+1+4+4 = 14 → s_d = √(14/4) = √3.5 ≈ 1.871.
• SE = 1.871/√5 = 1.871/2.236 ≈ 0.837. t = 5.0/0.837 ≈ 5.98, df = 4.
• Critical t at α = 0.05 two-tailed, df 4 = 2.776 → significant improvement. With n = 5 and no control group, maturation and testing effects remain plausible rival explanations.

Example 4 — sample size for a baseline survey. A district survey must estimate the proportion of households using a government health scheme, expecting p ≈ 0.5, margin of error 5 per cent, 95 per cent confidence; the universe is 10,000 households.

• n₀ = 1.96² × 0.5 × 0.5 / 0.05² = 3.8416 × 0.25 / 0.0025 = 0.9604 / 0.0025 = 384.2 ≈ 385.
• Finite population correction: n = 385 / [1 + (385 − 1)/10,000] = 385 / 1.0384 ≈ 370.8 ≈ 371 households.
• Allowing 10 per cent non-response: 371 / 0.90 ≈ 413 households to approach. Systematic sampling interval k = 10,000/413 ≈ 24.

Example 5 — Spearman's rank correlation. Six villages are ranked on number of SHGs (x) and on a women's participation score (y): d values are 0, 1, −1, 2, −1, −1.

• Σd² = 0 + 1 + 1 + 4 + 1 + 1 = 8; n = 6 → n(n² − 1) = 6 × 35 = 210.
• rho = 1 − (6 × 8)/210 = 1 − 48/210 = 1 − 0.2286 = +0.771 — a strong positive rank association between SHG density and participation, with no claim of causation.

Where it is used

• Monitoring and evaluation / research roles in NGOs and foundations — Pratham, Smile Foundation, SEWA, Azim Premji Foundation, Tata Trusts: baseline and endline surveys, KoBoToolbox tools, data cleaning and dashboards.
• Government and policy research — ICSSR institutes, NITI Aayog, state Social Welfare and Women & Child Development departments, social audits of welfare schemes, District Mental Health Programme studies.
• CSR teams and impact-assessment consultancies — needs assessment before a project and impact assessment required for large CSR projects under the Companies Act, 2013 rules; Sattva, NASSCOM Foundation-type assignments.
• Hospital and psychiatric social work — psychiatric social workers at NIMHANS, AIIMS and CIP Ranchi running single-system designs, scale administration and clinical audits with IEC clearance.
• Field action projects and academia — TISS field action projects, university dissertations, UGC-NET/JRF and Ph.D research; market and social research agencies for survey supervision.
• International agencies and research firms — UNICEF, UNDP, WHO India, J-PAL South Asia, IFMR LEAD: enumerator training, field supervision, randomised and quasi-experimental studies.

Interview questions

• Q: How does social work research differ from sociological research? A: The methods overlap, but social work research is driven by practice — it studies client needs, interventions and agency programmes with the aim of improving service and well-being, and is bound by professional ethics.
• Q: What is an operational definition and why does it matter? A: It states exactly how a concept will be measured — for instance, "empowerment" as a score on a 12-item decision-making scale — so that the study is replicable and the data are interpretable.
• Q: When do you use a schedule instead of a questionnaire? A: When respondents may have limited literacy or the questions need explanation; the investigator asks and records, which improves response rate and accuracy but costs more.
• Q: Which sampling method for a study of injecting drug users in a city? A: Snowball (respondent-driven) sampling, because there is no sampling frame; it is reported as non-probability, with the referral chains and their limits stated.
• Q: Difference between stratified and cluster sampling? A: Stratified samples from every stratum to cut sampling error when groups differ; cluster samples whole groups to cut cost and travel, usually at the price of a larger error.
• Q: Why do you test the null hypothesis rather than your own? A: A statistical test can only estimate the probability of the observed data under an assumption of no effect; rejecting that assumption is the evidence for the research hypothesis.
• Q: A tool is highly reliable. Is it valid? A: Not necessarily — it may consistently measure the wrong thing. Reliability is necessary but not sufficient for validity.
• Q: When would you report the median rather than the mean? A: For skewed or outlier-affected data such as income, land holding or treatment cost, and for ordinal data, where the mean is misleading.
• Q: Your chi-square is significant at p = 0.04 with n = 2,000. How much does that tell you? A: Only that the association is unlikely to be chance; with a large sample even a trivial association turns significant, so an effect size such as phi or Cramér's V must be reported.
• Q: What is a single-system design and when is it the right choice? A: Repeated measurement of one client or group across baseline and intervention phases (AB, ABAB, multiple baseline) — the practical way for a practitioner to evaluate their own casework when a control group is impossible.
• Q: How do you handle confidentiality when a respondent discloses child sexual abuse? A: Confidentiality yields to the law: POCSO, 2012 makes reporting mandatory. The consent form must state this limit in advance, and the child's identity must not be disclosed in the report.
• Q: Which Indian secondary sources would you use for a study on adolescent nutrition, and what would you check? A: NFHS-5, UDISE+, Poshan Tracker/ICDS and district statistical handbooks — checking definitions, reference period, sample design and base year before comparing across rounds.
• Q: How is rigour judged in a qualitative study? A: By credibility, transferability, dependability and confirmability — established through triangulation, member checking, an audit trail, thick description and reflexivity, not by p-values.

Common mistakes

• Applying a mean, t-test or Pearson's r to nominal or ordinal data (religion, caste, Likert categories treated as rupees).
• Writing the research design to fit a favourite tool or software instead of letting the objectives decide the design.
• Confusing universe with sampling frame, or calling a purposive/convenience sample "random" and then generalising to a district.
• Treating statistical significance as proof of importance or of causation, and omitting effect size and confidence intervals.
• Mixing up Type I and Type II errors, and forgetting that df = (r − 1)(c − 1) for chi-square, not n − 1.
• Skipping the pilot study and ethics clearance, or using a consent form in English for respondents who do not read it.
• Reporting percentages on very small bases (3 of 7 as "43 per cent") and dropping non-response from the denominator.

Related topics

• Monitoring and evaluation of development projects · Social work practice with individuals and families (casework) · Community organisation and development · Social policy and welfare administration · Social statistics and data analysis software · Project proposal and report writing.