Research Methodology and Methods: Philosophy of Science, Design, Methods and Techniques
2. Objectivity, reflexivity, and the ethics and politics of research
Objectivity is the demand that findings not be distorted by the researcher's wishes. Weber's answer is value-neutrality (Wertfreiheit) in analysis, while admitting that values decide which problems are worth studying. Gunnar Myrdal argued that complete value-freedom is impossible, so the honest course is to state one's value premises openly. Reflexivity turns the lens on the researcher: one's own class, gender, caste and standpoint shape what is seen and how respondents react, so a good account examines the investigator's position as part of the evidence rather than pretending to a view from nowhere.
3. Research design: fact, concept and theory; hypotheses, questions and objectives
A fact is a verified observation; a concept is an abstraction that names a class of things (alienation, caste, mobility); a theory is a set of logically linked propositions that explains a range of facts. A concept made measurable is a variable. A hypothesis is a tentative, testable statement of a relationship between variables — for example, that higher parental education raises a child's years of schooling. The independent variable is the presumed cause, the dependent variable the presumed effect; the null hypothesis states no relationship, and evidence is weighed against it.
A research question states what the study seeks to find out; objectives break it into specific, achievable aims; a hypothesis offers a provisional answer to be tested. Exploratory designs map a little-known area; descriptive designs portray a population accurately; explanatory designs test causal relations. Reliability is the consistency of a measure on repetition; validity is whether it measures what it claims — a measure can be reliable yet invalid (consistently wrong).
Reading social science research is itself a skill: find the research question, the concepts and how they were measured, the sample and how it was drawn, the method, and whether the conclusions follow from the evidence. Primary data are collected first-hand by the researcher; secondary data were collected by others (the Census, national sample surveys, administrative records). Documents such as letters, newspapers, reports and archives are weighed by John Scott's four criteria: authenticity (is it genuine and of known origin), credibility (is it free of error and distortion), representativeness (is it typical of the wider body of such documents) and meaning (is it clear and understandable in its own context). A source can be authentic yet unrepresentative, so one document is never a sample.
4. Quantitative and qualitative methods
| Method | What it does and its strength |
|---|---|
| Ethnography | Prolonged fieldwork and participant observation to grasp a group from within (emic); depth and context, but hard to generalise |
| Survey | Standardised data from a sample to describe a population and test relations; breadth and generalisability, but thin on meaning |
| Historical | Uses documents, archives and records to study the past and processes of change; interpretation of sources is central |
| Comparative | Compares cases or societies to find what varies with what; Durkheim treated comparison as the core sociological method |
| Statistical | Collects and analyses numerical data to summarise, correlate and infer; precise, but only as good as its measures |
Quantitative work seeks measurement, representativeness and generalisation; qualitative work seeks meaning, depth and context. They are complementary, and triangulation — using more than one method, source or observer on the same question — strengthens a finding by checking it from several angles.
Statistical methods are descriptive when they summarise the data in hand and inferential when they generalise from a sample to its population, by estimating a parameter with a confidence interval or by testing a hypothesis. A significance test asks how likely a result at least this extreme would be if the null hypothesis were true; if that probability, the p-value, falls below a chosen significance level such as 0.05, the null is rejected. The chi-square test examines association between two nominal variables in a cross-tabulation. Two traps: a p-value is not the probability that the null is true, and a statistically significant result is not necessarily a substantively important one.
5. Sampling
A sample is a subset drawn from a population listed in a sampling frame. The aim of probability sampling is representativeness: every unit has a known, non-zero chance of selection, which lets us estimate sampling error and generalise. The main designs are simple random (each unit equally likely), systematic (every k-th unit from a random start, where the interval k = population size ÷ sample size), stratified (divide the population into strata, then sample within each — in proportional allocation each stratum contributes in proportion to its share), and cluster/multistage (sample whole groups, then within them).
- Non-probability designs give no known selection probability and cannot measure sampling error, so they do not support statistical generalisation, though they are useful and economical: convenience (whoever is at hand), purposive/judgemental (chosen for a reason), quota (fill fixed numbers per category) and snowball (respondents refer others, useful for hidden populations).
- A larger sample reduces sampling error but not bias: a biased frame or a poor selection procedure gives a wrong answer however large the sample. The classic trap is confusing a big sample with a representative one.
6. Techniques of data collection: questionnaire, schedule, observation, interview, case study
- A questionnaire is filled in by the respondent, so it is cheap, wide-reaching and anonymous but fails with non-literate respondents and brings low response rates. A schedule holds the same questions but is filled in by a trained enumerator who asks and records, so it reaches non-literate respondents and clears up doubts, at higher cost.
- Observation may be participant (the researcher joins the group) or non-participant (watches from outside), and structured (a fixed schedule of what to record) or unstructured (open). Participant observation gives emic depth but risks the observer altering the scene and losing distance.
- An interview ranges from structured (fixed wording and order) through semi-structured to unstructured (a free conversation on themes). A case study examines a single unit — a person, family, village or event — in depth and in context; it is rich and holistic but cannot by itself establish how typical the case is.
7. Statistical analysis, interpretation and report writing
Analysis begins with the level of measurement: nominal (unordered categories, like religion), ordinal (ranked, like a Likert scale), interval (equal gaps, no true zero) and ratio (a true zero, like income), because the level fixes which statistics are legitimate. Central tendency is summarised by the mean (arithmetic average, for interval or ratio data, sensitive to outliers), the median (the middle value, good for skewed or ordinal data) and the mode (the most frequent value, the only one usable for nominal data). Dispersion is summarised by the range, the variance and the standard deviation.
Correlation measures how two variables move together (Pearson's r for interval/ratio data, Spearman's rank correlation for ranks), with r running from −1 through 0 to +1; a correlation is not a cause. Interpretation reads the numbers back into the research question and theory, acknowledging limits. A report states the problem, reviews the literature, describes the method and sample, presents the findings in tables and figures, discusses them against the hypotheses, and documents sources — written for a reader who was not in the field.
Key takeaways
- Popper's demarcation is falsifiability (science advances by conjecture and refutation, not induction); Kuhn describes normal science within a paradigm, anomalies, crisis and an incommensurable paradigm shift.
- Induction goes from particulars to a rule, deduction from a rule to particulars; emic uses insiders' categories and etic the analyst's; hermeneutics interprets meaning, and the double hermeneutic notes that the studied world is already interpreted.
- Objectivity seeks value-neutrality while making value premises explicit (Myrdal); reflexivity examines the researcher's own position; ethics demands informed consent, confidentiality and no harm, and research has a politics of funding and power.
- A concept made measurable is a variable; a hypothesis states a testable relation between an independent and a dependent variable; reliability is consistency and validity is measuring the right thing, and a measure can be reliable yet invalid.
- Probability sampling (simple random, systematic with k = N/n, stratified, cluster) permits generalisation; non-probability (convenience, purposive, quota, snowball) does not. A questionnaire is self-filled, a schedule is filled by an enumerator; a bigger sample cuts error but not bias.
- The level of measurement fixes the legitimate statistic: only the mode for nominal, the median suits ordinal or skewed data, the mean needs interval or ratio data; correlation runs from -1 to +1 and never by itself proves causation.
Practice questions (16)
Attempt each one before opening the answer. Every explanation names the tempting wrong option as well as the right one, because that is where marks are lost.
On Popper's view, what makes a claim scientific is that it is:
Show answer
Answer: B — falsifiable, forbidding some possible observation and so open in principle to refutation
For Popper the mark of science is falsifiability: a genuine theory rules something out and can therefore be tested and refuted. Accumulated confirmations do not demarcate science, because he denies that confirming instances prove a general law. A claim that fits any outcome is unfalsifiable, the opposite of scientific, and reliance on induction is exactly what he rejects in favour of conjecture and refutation.In Kuhn's account of science, a 'paradigm shift' is best described as:
Show answer
Answer: C — a revolutionary move to a new framework, prompted by anomalies and often incommensurable with the old
A paradigm shift is Kuhn's revolutionary break: anomalies pile up under normal science until a crisis forces a new framework, and the new paradigm is often incommensurable with the old. Steady cumulative addition and gradual refinement describe normal science, which a revolution interrupts; and 'final verification by a crucial experiment' is a naive verificationist picture that neither Kuhn nor Popper accepts, and it still misses his idea of incommensurability.A researcher observes many separate instances of a custom across villages and then formulates a general rule covering them all. This pattern of reasoning is:
Show answer
Answer: A — Inductive, moving from particular observed cases toward a general rule
Generalising from many particular observations to a rule is induction. Deduction runs the other way, from a general premise to a particular case. Falsificationism seeks refuting cases rather than building a generalisation, and the hermeneutic approach is about interpreting meaning, not inferring a general empirical rule.An anthropologist writes up a ritual strictly in the categories and meanings the participants themselves use, rather than in the comparative categories of outside analysis. This account is:
Show answer
Answer: C — An emic account, built from the insiders' own meanings and categories
Describing a practice in the insiders' own meanings is the emic standpoint. The etic standpoint is precisely the opposite, using the outside analyst's comparative categories. A nomothetic approach chases general laws, and a positivist one reduces the ritual to observable measurable behaviour, neither of which is the insider-meaning description specified.In the hypothesis "more years of parental schooling raise a child's years of schooling", the child's years of schooling is the:
Show answer
Answer: B — Dependent variable, taken to be the presumed effect in the stated relation
The variable treated as the effect, whose value is said to depend on another, is the dependent variable; here the child's schooling depends on the parents'. The parents' schooling is the independent variable (the presumed cause). A control variable is one held constant, and the null hypothesis is a statement of no relationship, neither of which fits the child's outcome.A team wants both the lived meaning of widowhood as understood by widows in one locality and a generalisable estimate of how widowhood affects incomes across a whole district. The most defensible design pairs:
Show answer
Answer: A — ethnography for the lived meaning and a sample survey for the district-wide estimate
Lived meaning calls for ethnography's in-depth, emic fieldwork, while a generalisable district estimate calls for a representative sample survey, so pairing the two fits each aim to its right method. Reversing them sends each method to the task it does badly. A lone case study cannot yield a district-wide generalisation, and the historical method studies the past from documents rather than recovering present lived meaning.A researcher stands outside one metro station and interviews whoever is willing to stop, then treats the results as representative of the whole city. The main methodological fault is that this is:
Show answer
Answer: C — convenience (non-probability) sampling, whose selection probabilities are simply unknown
Interviewing whoever happens to be available is convenience sampling, a non-probability method: there is no known probability of selection, so it cannot support statistical generalisation to the city. It is not stratified (no strata were defined) or systematic (no fixed interval from a random start), and it is certainly not simple random sampling, which would require every city resident to have an equal, known chance of selection.In the classic distinction used in Indian social research, the key difference between a questionnaire and a schedule is that:
Show answer
Answer: D — a questionnaire is filled in by the respondent, whereas a schedule is filled in by an enumerator
The defining difference is who records the answers: the respondent fills a questionnaire, while a trained enumerator administers and fills a schedule, which is why schedules can reach non-literate respondents. The other options invent differences: both can be written, both can carry open or closed items, and the questionnaire/schedule split is about mode of administration, not about being qualitative versus quantitative.A historian of rural debt relies on the account books preserved by one moneylending family and asks whether those books are typical of what other lenders in the region kept. In John Scott's four criteria for assessing documents, this question concerns the books':
Show answer
Answer: B — representativeness, that is, whether they are typical of the wider body of such records
Asking whether one family's books are typical of what other lenders kept is the representativeness criterion. Authenticity asks whether the books are genuine, which is not in doubt here. Credibility asks whether the entries are accurate and undistorted, and meaning asks whether they can be understood in their own context; the question raises neither, only typicality.A variable records religion as Hindu, Muslim, Christian, Sikh and other, with no inherent order. Which single measure of central tendency is legitimately usable for it?
Show answer
Answer: D — The mode, the most frequently occurring category
Religion is a nominal variable with unordered categories, and for nominal data only the mode is meaningful. The mean is illegitimate because the number codes are arbitrary labels, not quantities to average; the median needs a genuine order, which nominal categories lack; and the standard deviation is a measure of dispersion, not of central tendency, and also requires numeric data.A village register lists 1,500 households in order. A researcher wants a systematic random sample of 60 households. The sampling interval k (= population size / sample size) is ____.
Numerical answer — type the value.
Show answer
Answer: 25
In systematic sampling the interval k is the population size divided by the sample size: k = 1500 / 60 = 25. The researcher picks a random start between 1 and 25 and then takes every 25th household (start, start+25, start+50, ...) until 60 households are selected.A workforce of 4,000 has 2,500 in manufacturing and 1,500 in services. For a proportionally allocated stratified sample of 160 workers, the number to be drawn from services is ____.
Numerical answer — type the value.
Show answer
Answer: 60
Proportional allocation gives each stratum a share of the sample equal to its share of the population. Services are 1500 of 4000, a fraction of 0.375, so from a sample of 160 the services stratum contributes 0.375 x 160 = 60 (and manufacturing contributes the remaining 100).In a simple random sample of 400 respondents, exactly 50% say they trust local government, so the sample proportion is p = 0.5. The standard error of this proportion, computed as the square root of p(1 - p)/n and rounded to three decimal places, is ____.
Numerical answer — type the value.
Show answer
Answer: 0.025
The standard error of a proportion is the square root of p(1 - p)/n. Here p(1 - p) = 0.5 x 0.5 = 0.25, divided by n = 400 gives 0.000625, and its square root is 0.025 exactly. The common slip is to forget the square root and report 0.000625, or to divide by the square root of n at the wrong step.Which statements about objectivity, reliability and validity are correct?
Show answer
Answer: A — A measure can be reliable (consistent on repetition) yet invalid (consistently measuring the wrong thing); B — Reflexivity means examining how the researcher's own position shapes the research; C — Making one's value premises explicit is one honest response to the difficulty of full value-freedom
Reliability is consistency and validity is measuring the intended thing; a measure can be reliably wrong, so reliability does not guarantee validity. Reflexivity is exactly the scrutiny of the researcher's own influence, and stating value premises openly (Myrdal) is a recognised response to the limits of value-freedom. The claim that a highly reliable measure is thereby automatically valid is false, since a measure can reliably keep measuring the wrong thing.Which of the following correctly characterise the methods of social research?
Show answer
Answer: A — Ethnography uses prolonged fieldwork and participant observation to grasp a group from within; B — A sample survey trades some depth of meaning for breadth and the power to generalise to a population; C — Triangulation uses more than one method or source on the same question to strengthen a finding
Ethnography's prolonged participant fieldwork, the survey's trade of depth for generalisable breadth, and triangulation's cross-checking across methods are all accurate. The correlation claim is false: however strong, a correlation does not establish causation, since a third variable may drive both or the direction may be reversed.Which statements about the philosophy and ethics of social research are correct?
Show answer
Answer: A — Voluntary informed consent and confidentiality are core ethical obligations toward those studied; B — The double hermeneutic notes that social science interprets a world already interpreted by its members; C — The politics of research concerns who funds it, whose questions are asked and who benefits
Informed consent and confidentiality are basic research ethics; the double hermeneutic (Giddens) captures the interpreted-world point; and funding, agenda and benefit are the politics of research. The claim that a larger sample removes bias from a flawed frame is false: increasing sample size reduces random sampling error but not the systematic bias built into a flawed frame, which persists at any sample size.