14:53
003-experimental Design
Ian Walters
Overview
This video introduces fundamental terminology and concepts in experimental design, focusing on how researchers manipulate variables to study cause-and-effect relationships. It defines key terms like experimental unit, characteristic, factor, factor level, and treatment, illustrating them with examples from a drug study and an agricultural study. The video then contrasts two primary methods for assigning experimental units to treatments: completely randomized designs and randomized block designs, explaining how each approach aims to control for extraneous variables and ensure valid conclusions.
How was this?
Save this permanently with flashcards, quizzes, and AI chat
Chapters
- An experimental unit is the object or entity on which a study is performed (e.g., a person, a plot of land).
- A characteristic is a property of an experimental unit that cannot be manipulated by the researcher (e.g., gender, prior soil quality).
- A factor is a variable that the researcher actively manipulates or changes.
- Factor levels are the specific values or categories of a factor that are used in the study.
- Treatments are the specific conditions or combinations of factor levels applied to experimental units; in a one-factor study, treatments are the factor levels themselves, while in a multi-factor study, treatments are combinations of factor levels.
Understanding these terms is crucial for clearly defining the scope and methodology of any experiment, ensuring that the research question can be precisely addressed.
In an antibiotic study, people are the experimental units, dosage level is the factor, and 3ml, 5ml, 7ml, 10ml, and 15ml are the factor levels, which also serve as the treatments in this one-factor study.
- The antibiotic study used people as experimental units, with dosage as the manipulated factor and specific milliliters as factor levels/treatments.
- The agriculture study used plots of land as experimental units, with seed type and fertilizer type as two distinct factors.
- In the agriculture study, prior soil quality was a characteristic that could not be manipulated but might influence results.
- For the multi-factor agriculture study, treatments were combinations of seed types and fertilizer types (e.g., Seed 1 + Fertilizer A).
These examples demonstrate how the abstract definitions of experimental units, characteristics, factors, levels, and treatments apply to real-world research scenarios, solidifying comprehension.
In the agriculture study, a treatment could be 'Seed Type S1' combined with 'Fertilizer Type F1', representing one specific experimental condition applied to a plot of land.
- Experimental design focuses on how to assign experimental units to different treatment groups once a sample is obtained.
- A completely randomized design involves randomly assigning experimental units to all available treatment groups.
- This random assignment relies on chance to distribute characteristics evenly across groups, hoping to control for extraneous variables.
- A randomized block design involves first grouping experimental units into 'blocks' based on a characteristic that might affect the outcome.
- Within each block, experimental units are then randomly assigned to the treatment groups, ensuring that each treatment group has an equal representation of units from that block.
The method of assigning units to treatments directly impacts the validity of the study's conclusions by influencing how well potential confounding variables are controlled.
In a flipped classroom study, a completely randomized design would randomly assign all students to either the traditional or flipped class, while a randomized block design would first group students by SAT math scores and then randomly assign students from each score group to both teaching methods.
- Randomized block designs offer more control than completely randomized designs by proactively addressing known sources of variation.
- Blocking uses a characteristic (like SAT math score) that cannot be manipulated but is believed to influence the outcome variable.
- By ensuring equal representation of different levels of the blocking characteristic across all treatment groups, the design controls for its potential impact.
- This method allows researchers to isolate the effect of the treatment more effectively, as the influence of the blocked characteristic is balanced across groups.
Randomized block designs strengthen experimental control, leading to more precise and reliable findings by mitigating the influence of specific, known extraneous variables.
Grouping students by SAT math score (e.g., 700+, 600-700, 500-600, <500) and then randomly assigning half of each group to the traditional class and half to the flipped class ensures that differences in math ability are equally represented in both teaching method groups.
Key takeaways
- Experimental units are the subjects of study, while factors are variables manipulated by the researcher.
- Factor levels are the specific values of a factor used, and treatments are the conditions applied to experimental units.
- Multi-factor studies involve combinations of factor levels to create treatments.
- Completely randomized designs rely on chance to balance characteristics across treatment groups.
- Randomized block designs proactively control for known sources of variation by grouping units before random assignment.
- Blocking ensures that characteristics believed to influence outcomes are equally represented across all treatment groups.
- Effective experimental design is crucial for establishing causal relationships and drawing valid conclusions.
Key terms
Experimental UnitCharacteristicFactorFactor LevelTreatmentCompletely Randomized DesignRandomized Block DesignBlock
Test your understanding
- What is the difference between a factor and a characteristic in an experiment?
- How are treatments defined in a one-factor study versus a multi-factor study?
- Why would a researcher choose a randomized block design over a completely randomized design?
- How does blocking help control for extraneous variables in an experiment?
- What is the role of randomization in both completely randomized and randomized block designs?