3.2 Steps of Research & Experimental Methodologies

Steps of Research & Experimental Methodologies

Conducting rigorous academic research requires mastery of a structured methodological pipeline, sound experimental controls, appropriate sampling frameworks, and formal statistical hypothesis testing procedures.

---

1. The Step-by-Step Research Pipeline

The scientific research process follows a canonical sequential pipeline, where each phase establishes the prerequisite foundation for subsequent steps:

┌───────────────────────────────────┐
│ 1. Formulate the Research Problem │ (Define scope, background & core question)
└─────────────────┬─────────────────┘
                  ▼
┌───────────────────────────────────┐
│ 2. Extensive Literature Review    │ (Identify knowledge gaps & conceptual models)
└─────────────────┬─────────────────┘
                  ▼
┌───────────────────────────────────┐
│ 3. Formulate Testable Hypotheses  │ (Null H0 vs Alternative H1; Directional/Non-directional)
└─────────────────┬─────────────────┘
                  ▼
┌───────────────────────────────────┐
│ 4. Develop the Research Design    │ (Blueprint: Experimental, Descriptive, or Correlational)
└─────────────────┬─────────────────┘
                  ▼
┌───────────────────────────────────┐
│ 5. Select Sampling Design         │ (Probability vs Non-Probability; Determine Sample Size)
└─────────────────┬─────────────────┘
                  ▼
┌───────────────────────────────────┐
│ 6. Collect Empirical Data         │ (Questionnaires, Structured Schedules, Observational Tools)
└─────────────────┬─────────────────┘
                  ▼
┌───────────────────────────────────┐
│ 7. Analyze & Process Data         │ (Descriptive stats, Inferential testing: t-test, ANOVA, Chi-Square)
└─────────────────┬─────────────────┘
                  ▼
┌───────────────────────────────────┐
│ 8. Test Hypotheses & Interpret    │ (Retain/Reject H0, Relate to theoretical literature)
└─────────────────┬─────────────────┘
                  ▼
┌───────────────────────────────────┐
│ 9. Prepare the Research Report    │ (Format thesis chapters, document limitations & implications)
└───────────────────────────────────┘

---

2. Experimental Research & Variable Taxonomy

Experimental research is the gold standard for establishing cause-and-effect relationships through active manipulation, control, and random assignment.

Classification of Variables

• Independent Variable (IV / Treatment / Predictor): The antecedent condition or variable actively manipulated by the researcher to observe its effect (e.g., Computer-Assisted Instruction vs. Traditional Lecture).
• Dependent Variable (DV / Criterion / Outcome): The consequential variable that is measured and observed to change in response to IV manipulation (e.g., Student Academic Achievement Scores).
• Extraneous Variables: Uncontrolled variables not central to the study that could inadvertently influence the dependent variable (e.g., room temperature, baseline IQ, socioeconomic background).
• Confounding Variables: An extraneous variable that covaries systematically with the independent variable, making it impossible to disentangle which variable caused the observed change in the dependent variable.
• Moderating Variable: A secondary independent variable that modifies the direction or strength of the relationship between the IV and DV (e.g., student gender moderating the effect of instructional method on performance).
• Mediating (Intervening) Variable: An internal explanatory mechanism or process through which the independent variable influences the dependent variable (e.g., instructional method increases student motivation, which in turn elevates achievement).

Control Mechanisms in Experiments

1. Random Assignment (Randomization): Equates experimental and control groups across measured and unmeasured extraneous characteristics before treatment.
2. Matching: Pairing subjects with identical attributes (e.g., matched IQ scores) across treatment arms.
3. Statistical Control: Utilizing covariate adjustments such as Analysis of Covariance (ANCOVA) to partial out extraneous variance mathematically.

Threats to Experimental Validity

Quasi-Experimental Designs

When true randomization is ethically or practically unfeasible (such as using intact school classrooms), researchers employ Quasi-Experimental Designs (e.g., Non-Equivalent Control Group Pretest-Posttest Design or Interrupted Time-Series Design), controlling for confounds through statistical modeling.

---

3. Sampling Techniques: Probability vs. Non-Probability

Sampling is the process of selecting a representative subset of units from a defined target population to make valid statistical generalizations.

                                   SAMPLING TECHNIQUES
                                            │
            ┌───────────────────────────────┴───────────────────────────────┐
            ▼                                                               ▼
   PROBABILITY SAMPLING                                           NON-PROBABILITY SAMPLING
 (Every unit has known, non-zero P)                              (Subjective, Non-random selection)
            │                                                               │
 ┌──────────┼──────────┬──────────┐                      ┌──────────┬───────┼──────────┐
 ▼          ▼          ▼          ▼                      ▼          ▼       ▼          ▼
Simple   Systematic Stratified  Cluster               Purposive   Quota  Snowball  Convenience
Random   (k = N/n)  (Homog.   (Heterog.              (Judgment) (Non-    (Chain-   (Accidental)
                     strata)   clusters)                         random  referral)
                                                                 strata)

Probability Sampling Methods

1. Simple Random Sampling: Every member of the population has an equal, independent probability of selection (lottery method or random number generators).
2. Systematic Sampling: Selecting units at uniform intervals from an ordered sampling frame using the sampling fraction $k = N / n$, where $N$ is total population size and $n$ is desired sample size, starting from a randomly determined starting point between $1$ and $k$.
3. Stratified Random Sampling: The heterogeneous population is divided into mutually exclusive, internally homogeneous subgroups (strata) based on key characteristics (e.g., rural vs. urban, income levels), after which simple random samples are drawn from each stratum (proportionate or disproportionate).
4. Cluster Sampling: The population is naturally divided into internally heterogeneous groups (clusters, e.g., entire schools or geographical districts). Complete clusters are randomly sampled, and all individuals within chosen clusters are surveyed (often executed as Multi-stage Cluster Sampling).

Non-Probability Sampling Methods

1. Purposive / Judgmental Sampling: Researchers hand-pick specific information-rich cases based on expert judgment and defined study criteria.
2. Quota Sampling: Sets demographic quotas (e.g., 50% male, 50% female) mirroring population proportions, but units are selected conveniently rather than randomly.
3. Snowball (Chain-Referral) Sampling: Initial participants recruit subsequent eligible peers from their social network, ideal for hidden, marginalized, or hard-to-reach populations (e.g., rare disease patients, underground artists).
4. Convenience / Accidental Sampling: Selecting readily available and accessible participants (e.g., surveying students in the university cafeteria).

---

4. Hypothesis Testing & Decision Matrix (Type I vs. Type II Errors)

In statistical hypothesis testing, researchers evaluate empirical evidence against a default baseline assertion—the Null Hypothesis ($H_0$), which posits no true difference or relationship—compared to the Alternative Hypothesis ($H_1$).

The Four-Quadrant Decision Matrix

• Level of Significance ($\alpha$): The probability of committing a Type I error, conventionally set a priori at $0.05$ (5%) or $0.01$ (1%).
• Statistical Power ($1 - \beta$): The probability of correctly rejecting a false null hypothesis. Power is enhanced by increasing sample size ($n$), increasing effect size, or lowering measurement error.