Exhaustive Study Notes on Experimental Design and Statistical Methods

Blocking in Experimental Design

  • Core Concept: Blocking is one of the four foundational principles of experimental design. While not strictly mandatory in every experiment, it is strongly recommended whenever experimental units share a known, pre-existing similarity or characteristic that could influence the response variable.
  • Mechanism of Blocking:
    • Experimental units are grouped into homogeneous blocks based on shared attributes prior to treatment administration.
    • Within each individual block, experimental units are randomly assigned across treatment groups.
  • Example: Chlorine Effect on Acrylic Nails:
    • Objective: Evaluate how immersion in chlorinated pool water affects nail polish chipping on acrylic nails.
    • Experimental Treatments: Red nail polish versus Nude nail polish.
    • Source of Nuisance Variability: Hand dominance and daily hand usage patterns. A subject's dominant hand undergoes more daily wear and tear, leading to higher baseline chipping regardless of polish color.
    • Blocking Strategy: Block by individual hands (Left Hand vs. Right Hand).
    • Procedure: Exclude thumbs, leaving 88 non-thumb fingers (44 on the left hand, 44 on the right hand). For the left hand block, randomly assign 22 fingers to receive red nail polish and 22 fingers to receive nude nail polish. For the right hand block, randomly assign 22 fingers to receive red nail polish and 22 fingers to receive nude nail polish.
    • Purpose: Prevents one polish color from being overrepresented on the dominant hand. It ensures each hand serves as its own internal control.
    • Parallel Concept: Blocking in experimental design is conceptually identical to stratification in sampling and observational study design.
    • Simultaneous Replication: Treating each hand as an independent block replicates the experiment concurrently within every subject.
    • Results of Experiment: Nude nail polish exhibited a higher overall percentage of chipping compared to red nail polish, contradicting initial assumptions that bolder colors chip faster.
  • Formal Definitions:
    • Blocks: Groups of experimental units created based on non-treatment attributes shared prior to treatment.
    • Blocking: The procedure of randomly assigning treatments within each individual block to isolate variability caused by differences between blocks.
    • Randomized Block Design: An experimental structure in which units are divided into blocks before being randomly assigned to treatment conditions within each block.

Completely Randomized Design

  • Definition: A completely randomized design is the simplest experimental design structure. Experimental units are assigned directly to treatment groups using a purely random mechanism without any prior blocking or stratification based on baseline characteristics.
  • Comparison to Randomized Block Design:
    • Completely Randomized Design: 3030 acrylic nails →\rightarrow Random Assignment →\rightarrow 1515 nails to Red Polish, 1515 nails to Nude Polish →\rightarrow Compare percent chipped.
    • Randomized Block Design: 88 fingers →\rightarrow Divide into Left Hand Block (44 fingers) and Right Hand Block (44 fingers) →\rightarrow Randomly assign 22 fingers to Red and 22 to Nude within each hand →\rightarrow Compare percent chipped within and across hands.

Matched Pairs Design

  • Definition: A matched pairs design is a specialized form of a randomized block design where block sizes are restricted to exactly two (22) highly similar experimental units, or where a single experimental unit receives both treatments sequentially.
  • Procedure: Subjects are paired based on critical blocking variables (e.g., exact age, gender, twin status). Within each pair, the two subjects are randomly assigned to opposing treatment conditions.
  • Hypothetical Medical Experiment Example:
    • Population: 10001000 subjects.
    • Pairing: Form 500500 matched pairs using gender and age as matching factors.
    • Pair 11: Two 2121-year-old women.
    • Pair 22: Two 2121-year-old men.
    • Pair 33: Two 2222-year-old women.
    • Assignment: Within each pair, one subject is randomly assigned to receive a cold vaccine, and the other subject receives a placebo (e.g., saline injection).
  • Rationale: Controls for high degrees of subject-to-subject variation, allowing researchers to evaluate treatment differences as if the same baseline profile received both treatments.

Pet Food Safety Experiment (Cats and Dogs)

  • Context: Testing the safety of a questioned pet food brand regarding kidney failure risks in both cats and dogs.
  • Subjects: 3030 cats and 4040 dogs.
  • Design Choice: Randomized Block Design (blocked by species to account for baseline biological differences between cats and dogs).
  • Experimental Diagram & Steps:
    • Initial group: 3030 cats and 4040 dogs.
    • Block 11 (3030 Cats) →\rightarrow Random Assignment →\rightarrow 1515 Cats given Safe Food, 1515 Cats given Questionable Food →\rightarrow Compare kidney function.
    • Block 22 (4040 Dogs) →\rightarrow Random Assignment →\rightarrow 2020 Dogs given Safe Food, 2020 Dogs given Questionable Food →\rightarrow Compare kidney function.
  • Random Assignment Implementation via Random Number Generator:
    1. Sequentially label all cats with unique integers from 11 to 3030.
    2. Execute a random number generator to select 1515 unique integers between 11 and 30$.\n 3. Assign the 15 cats corresponding to those selected numbers to the safe food group.\n 4. Assign the remaining 15 cats to the questionable food group.\n 5. Repeat the labeling (1toto40)anduniquerandomnumbergenerationprocess() and unique random number generation process (20 unique selections) for the dog block.\n\n# Confounding and Lurking Variables\n\n* **Lurking Variable**: A variable not included in an observational study that influences both the explanatory and response variables, creating a false impression of a direct causal link.\n* **Confounding Variable**: An extraneous factor in an experiment whose effect on the response variable cannot be distinguished from the effect of the explanatory variable.\n* **Example: Professor Teaching Style Study**:\n * Fall Semester: Subdued teaching demeanor.\n * Spring Semester: High enthusiasm and animated gestures.\n * Baseline Controls: Identical assignments, notes, and exam structures across both semesters.\n * Outcome: Spring end-of-semester learning evaluation ratings (scale 1toto5)averagedover) averaged over1 point higher than Fall evaluations.\n * Potential Confounding Variables:\n * Seasonal/Weather Differences: Pleasant spring weather versus gloomy fall weather affecting student mood.\n * Academic Maturity: Fall students adjusting to college baseline versus Spring students accustomed to academic rigor.\n * Implication: High ratings cannot be definitively attributed to animated teaching because weather and maturity were confounded with the teaching style.\n* **Example: Bank Offer Design (2 Factors)**:\n * Initial Setup: 50{,}000low−rate/no−feeoffersversuslow-rate/no-fee offers versus50{,}000 high-rate/with-fee offers.\n * Problem: Interest rate level and fee presence are completely confounded; customer preference cannot be isolated to the low interest rate or the absence of fees.\n * Unconfounded 2 \times 2 Factorial Design:\n 1. Low interest rate with no fee.\n 2. Low interest rate with fee.\n 3. High interest rate with no fee.\n 4. High interest rate with fee.\n\n# Concept Check and Discussion Questions\n\n* **Identification of Response Variable**:\n * **Scenario**: Determining optimal parameter settings for tile firing to produce tiles with high flexural strength.\n * **Explanatory Variables**: Firing temperature and firing time.\n * **Response (Dependent) Variable**: Flexural strength of the tiles.\n* **Calculating Treatment Combinations**:\n * **Scenario**: Experiment evaluating 3firingtemperaturesettingsandfiring temperature settings and4 firing time settings.\n * **Calculation**: \text{Total Treatments} = 3 \times 4 = 12 unique treatment combinations.\n* **Critical Components of Experimental Design**:\n * **Essential Principles**: Control, Replication, and Random Assignment.\n * **Non-Essential Element**: **Random Selection** of subjects from the population is NOT mandatory for a valid experiment. While random selection aids generalizability (common in observational sampling), experimental studies rely on **Random Assignment** to establish causal relationships, often using volunteer subjects.\n* **Blinding Classifications**:\n * **Scenario**: Subjects receive 3$$ unlabeled bottles (two containing active aspirin and one containing placebo). Subjects know one bottle is a placebo but cannot identify which. Testing center staff administering and evaluating the bottles also lack information regarding bottle contents.
    • Classification: Double-Blind Experiment.
    • Rationale: Neither the subjects receiving the treatments nor the evaluators administering the study know which specific treatment is contained in any given bottle.

Exam Review Overview

  • Chapter 1 Focus:
    • Categorical versus Quantitative variable classification.
    • Contextual study identification: Who (subjects/units), Why (purpose), and How (data collection method).
  • Chapter 2 Focus:
    • Graphical representation distributions for categorical and quantitative data.
    • Manual and calculator-based calculation of summary statistics.