Learn — Study designs & qualitative research

5 notes · 15 questions on the topic

Research designs — experimental, quasi-experimental and non-experimental

A research design is the overall plan for answering the research question — whether there is an intervention and who receives it, how the groups are formed, how many times data are collected and how bias is controlled. Quantitative designs fall into three families by the amount of control the researcher has — experimental, quasi-experimental and non-experimental — and the family a study belongs to decides how far its results can claim cause and effect.

A true experimental design has three properties together: manipulation (the researcher introduces the intervention), control (a comparison group that does not receive it) and randomisation (participants are assigned to the groups by chance); the randomised controlled trial is its clinical form and the standard for proving cause and effect.
A quasi-experimental design has manipulation but lacks randomisation, a control group or both — the non-equivalent control group design (two intact wards, one with the new protocol), the time-series design (repeated measurements before and after on one group), the one-group pre-test–post-test design; it is used where random assignment is impossible or unethical, and its weakness is that the groups may differ from the start.
A non-experimental (observational) design has no manipulation at all — descriptive designs that describe what exists, correlational designs that relate variables as they occur, comparative designs, retrospective (ex post facto, case-control) and prospective (cohort) designs — and it cannot prove cause because the presumed cause was never controlled.
Time gives a second axis: a cross-sectional design collects data once, at one point in time, and a longitudinal design collects data from the same people at several points (panel, trend, follow-up studies); a retrospective design looks back from the outcome to its cause and a prospective design looks forward from the cause to the outcome.
The levels of evidence rank the designs — level I a systematic review or meta-analysis of randomised trials, II a well-designed randomised trial, III a controlled trial without randomisation, IV a case-control or cohort study, V a systematic review of descriptive or qualitative studies, VI a single descriptive or qualitative study and VII the opinion of experts — the pyramid every evidence-based practice question draws.
Internal validity asks whether the intervention and nothing else caused the outcome and is threatened by history, maturation, testing, instrumentation, selection, attrition (mortality) and regression to the mean; external validity asks whether the findings generalise beyond the study; the Hawthorne effect, blinding and a placebo control are the words that go with them.
The choice follows the question: to prove cause, experiment; where randomisation cannot be done, quasi-experiment; to describe or relate, non-experimental; to understand meaning, qualitative — and the report must name the design so that the reader knows what the findings can and cannot claim.

True experimental design — randomisation, manipulation, control and the RCT

A true experimental design is one in which the researcher manipulates an independent variable, controls the situation with at least one comparison group and assigns participants to the groups at random, so that a difference in the outcome can be attributed to the intervention and not to who was in which group. The randomised controlled trial is its clinical form, and the designs examinations ask are written in the notation R, O and X — random assignment, observation and treatment.

The three essentials are manipulation — the intervention (X) given to the experimental group — control — a group that gets usual care, a placebo or nothing — and randomisation — every participant has an equal chance of landing in any group, done with a random number table, a computer or sealed opaque envelopes, never by alternate allocation, the day of the week or the time of admission.
The classic designs are the pre-test–post-test control group design (both groups measured before and after, only one treated), the post-test-only control group design (measured after only), the Solomon four-group design (which adds two groups without a pre-test to detect the effect of the pre-test itself), the factorial design (two or more independent variables tested at once, such as 2 × 2), the crossover design (each participant receives both treatments in random order and serves as her own control) and cluster randomisation (whole wards or villages are randomised rather than individuals).
Random assignment — to the groups — makes the groups comparable at the start on every known and unknown variable, and that is what lets an experiment claim cause; it is not random sampling, which is selection from the population and serves generalisation, and a trial can have one without the other.
Bias is kept out by blinding — single (the participant), double (the participant and the person giving or assessing the treatment), triple (the analyst as well) — by concealing the allocation until the moment of assignment, by a placebo, and by intention-to-treat analysis in which every participant is analysed in the group she was assigned to; the CONSORT statement is the reporting standard.
Drug trials pass through phases — I for safety and dose in a few volunteers, II for efficacy and side effects in a larger group, III the large randomised comparison against the standard treatment, IV post-marketing surveillance — and the phase-III trial is the randomised controlled trial of the textbooks.
The design's strength is the strongest possible evidence for cause; its weaknesses are cost, artificiality, the Hawthorne effect, non-compliance and attrition, the difficulty of blinding a nursing intervention, and ethics — no one can be randomised to smoke, which is why some questions are left to cohort studies.

Descriptive, correlational, cross-sectional and longitudinal designs

Non-experimental designs observe variables as they occur, without any intervention by the researcher, and they are the commonest designs in nursing research because most of the variables nurses study — age, disease, attitude, workload — cannot be manipulated. They describe what exists, relate one variable to another and follow change over time, and their limit is that they cannot prove cause.

A descriptive design describes the characteristics, frequency or distribution of a phenomenon — how many nurses wash their hands correctly, the prevalence of pressure injuries — with no hypothesis about relationships and no intervention; the descriptive survey, which collects self-report data by questionnaire or interview, is its usual form.
A correlational design examines the relationship between two or more variables without manipulating either — anxiety scores and hours of sleep in the same patients — and reports a correlation coefficient; correlation is not causation, because a third variable may explain both and the direction of influence is unknown.
A cross-sectional design collects data from a population at one single point in time — quick, cheap and the right design for prevalence — but it cannot separate cause from effect or show change; a longitudinal design collects data from the same people at several points — panel, trend and follow-up studies — and shows change and sequence at the price of time, money and attrition.
A retrospective (ex post facto) design starts from an outcome and looks back for the exposure, a prospective design starts from the exposure and follows forward, and a comparative descriptive design compares groups that already differ, such as smokers and non-smokers.
Survey research — by mailed or online questionnaire, by personal or telephone interview — is the method behind most descriptive and correlational designs, and its quality rests on the sampling and the response rate; case studies describe one unit in depth, secondary analysis and meta-analysis reuse existing data, and a methodological study develops a tool.
In epidemiological terms a cross-sectional survey gives the prevalence of a condition and a cohort study its incidence, and a stem that describes one measurement of two variables in one group with no intervention is naming a correlational, cross-sectional design.

Cohort and case-control studies — the epidemiological designs

Cohort and case-control studies are the two analytical observational designs of epidemiology that test whether an exposure is associated with a disease without the researcher assigning the exposure. A cohort study starts with the exposure and follows people forward to see who develops the outcome; a case-control study starts with the outcome and looks back to compare past exposure, and that difference decides which measure of risk each can give.

A cohort (prospective, follow-up, incidence) study takes a group free of the disease, classifies its members by exposure — smokers and non-smokers — and follows them over time; it gives the incidence in each group and from it the relative risk (incidence in the exposed divided by incidence in the unexposed) and the attributable risk, it can study several outcomes of one exposure, and its costs are money, years, unsuitability for rare diseases and loss to follow-up; a historical (retrospective) cohort reconstructs the follow-up from old records.
A case-control (retrospective) study takes people who already have the disease (cases), matches them with similar people who do not (controls), and looks back at past exposure in each; it gives an odds ratio, which estimates the relative risk when the disease is rare, but no incidence; it is quick, cheap, the right design for a rare disease or a long latency and can study many exposures for one outcome, and its weaknesses are recall bias, the choice of controls and uncertainty about which came first.
Direction is the memory aid: a cohort study runs from cause to effect, a case-control study from effect to cause, and a cross-sectional study measures both at once and gives prevalence.
Both are analysed in a 2 × 2 table of exposed and unexposed against diseased and not diseased — a, b, c, d — with relative risk = [a ÷ (a + b)] ÷ [c ÷ (c + d)] and odds ratio = ad ÷ bc; a value of 1 means no association, above 1 a risk, below 1 protection, and a confidence interval that includes 1 means the association is not statistically significant.
Bias has its home in each design — recall bias and selection of controls in the case-control study, loss to follow-up in the cohort — and confounding, a third factor linked to both exposure and disease, is handled by matching, stratification or multivariate analysis; Bradford Hill's criteria for judging causation (strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, analogy) count temporality as the only essential one.
On the ladder of evidence for cause the randomised trial stands above the cohort study, the cohort above the case-control, and the case-control above the cross-sectional survey, the case series and the single case report; Doll and Hill's studies of smoking and lung cancer — the 1950 case-control study and the British doctors' cohort — are the classic pair.

Qualitative research — phenomenology, grounded theory and ethnography

Qualitative research is the naturalistic, inductive inquiry into the meaning people give to their experiences, collected as words rather than numbers through interviews, observation and documents and analysed for themes; it answers questions of what an experience is like and how a process unfolds, which no scale can measure. Its three classic traditions — phenomenology, grounded theory and ethnography — are told apart by the question each asks and the discipline each grew from.

Phenomenology, from philosophy (Husserl's descriptive and Heidegger's interpretive or hermeneutic forms), describes the lived experience of a phenomenon — living with chronic pain, becoming a mother — through in-depth interviews with a small purposive sample, with the researcher bracketing her own beliefs and analysing the transcripts (Colaizzi, Giorgi, van Manen) into the essence of the experience.
Grounded theory, from sociology (Glaser and Strauss, 1967), aims to generate a theory grounded in data collected from participants about a social process — how families adapt after a stroke — by constant comparative analysis, theoretical sampling, coding in stages (open, axial, selective) until saturation, and the naming of a core category or basic social process.
Ethnography, from anthropology (and Leininger's ethnonursing in nursing), describes a culture or subculture from within — the culture of an intensive care unit, a community's birth practices — through months of fieldwork with participant observation and key informants, distinguishing the emic (insider) from the etic (outsider) view.
The other approaches are historical research on the documents of the past (primary and secondary sources, external and internal criticism), the case study, action research in which researchers and participants change practice together through cycles of plan, act, observe and reflect, narrative inquiry, and focus groups as a method of data collection.
Qualitative studies share their characteristics — the natural setting, the researcher as the instrument, an emergent and flexible design, purposive sampling that stops at data saturation (when no new information appears), inductive thematic analysis, thick description and reflexivity — and their samples are small and never claim to be statistically representative.
Rigour is called trustworthiness (Lincoln and Guba): credibility through prolonged engagement, member checking and triangulation of data, methods or investigators; dependability through an audit trail; confirmability; transferability through thick description; and authenticity — the qualitative counterparts of validity and reliability.

Free today: the 10-question daily quiz · the NORCET 11 recall paper · what ₹999 unlocks — once, no auto-renewal.