CRIM_220_-_Lecture_6_-_QEs_and_Sampling

Lecture Overview

Date: February 27, 2025Course: CRIM 220: Research Methods in CriminologyInstructor: Chelsey Lee

Quasi-Experimental Designs

DefinitionQuasi-experimental designs are research methods that involve comparisons between groups that are not the result of random assignment, leading to challenges in establishing causal relationships. They are utilized in settings where random assignment is not ethical or feasible. Common designs include non-equivalent groups and time-series designs, both of which aim to infer causal relationships despite these limitations.

Key Characteristics

• Non-random assignment can lead to group differences that impact the study's findings, raising concerns about internal validity.

• The absence of randomization often results in non-equivalent groups, making it harder to control for confounding variables.

Building Blocks of Quasi-Experiments

• Number of Groups: Studies typically include treatment and comparison groups, both of which should ideally be similar in all essential characteristics except for the treatment.

• Variations of IV: Researchers may investigate different levels or types of independent variables, allowing for a richer understanding of the treatment’s impact.

• Pretest & Posttest Measures: Data is collected before and after the treatment, allowing researchers to assess the effects of the intervention.

• Participant Selection & Assignment Procedures: Non-random methods such as matching or controlled selection are employed to assign participants to their respective groups, meaning researchers must explicitly account for their selection criteria.

Non-Equivalent Groups Designs

Characteristics

• The key feature is the lack of randomization leading to potential biases and non-equivalence of groups which may confound the results.

• The control group plays a critical role as a standard for comparison against the treatment group, with efforts made to minimize initial differences through matching techniques.

Types of Non-Equivalent Designs:

1. Static-Group Comparison:

   • A quasi-equivalent approach where treatment leads to potential group differences observed in the findings.

   • Structure: Group A: Treatment (X) → Posttest (0), Group B: No Treatment (0)

2. Pretest-Posttest Nonequivalent Control Group:

   • Measures are taken before and after treatment to assess potential group differences.

   • Structure: Group A: 0 → X → 0, Group B: 0 → 0

   • This design protects against some internal validity threats by providing baseline measurements.

3. One Group Pretest-Posttest:

   • Studies a single group before and after treatment, demonstrating change but lacking the ability to definitively establish causation.

   • Structure: Group A: Pretest (0) → Treatment (X) → Posttest (0)

Time-Series Designs

Characteristics

• A form of longitudinal research where multiple observations are made over time to observe trends and changes.

• Internal validity concerns arise particularly from threats posed by instrumentation or historical events.

Types of Time-Series Designs:

1. Interrupted Time-Series:

   • Involves multiple measures taken before and after treatment, which provides a clearer picture of the treatment's effects.

   • Structure: Group A: Treatment (X) → Multiple Posttests (0 after treatment)

2. Interrupted Time-Series with Non-Equivalent Comparison Group:

   • Incorporates an additional comparison group to control for external influences on the observed effects.

   • Structure: Group A: Treatment (X) → Posttests, Group B: No Treatment (0 posttests)

3. Interrupted Time-Series with Removed Treatment:

   • Analyzes the effects of both adding and later removing treatment, providing insight into the dependency of effects on the treatment condition.

   • Structure: Group A: Treatment → No Treatment

Non-Experimental Designs

Characteristics

• These designs do not manipulate the independent variable, lack a comparison group, and do not include repeated measures.

• Common forms include cross-sectional surveys and observational studies, which provide rich context but less control over external variables.

Sampling Overview

Purpose of SamplingSampling involves selecting observations from a larger population to draw conclusions and generalize findings. The primary aims are:

1. To enhance the feasibility of data collection within resource constraints.

2. To ensure a representative and relevant sample from the population being studied.

Types of Sampling

1. Probability Sampling: Each member of the population has an equal chance of being selected, which enhances the generalizability of results.

2. Non-Probability Sampling: Selection chances are not known, often limiting the representativeness of the sample and introducing bias.

Key Components of a Sample:

• Sample Element: The specific unit being studied within the broader population (e.g., individuals, institutions).

• Population: The entire group of individuals or instances that a researcher is interested in studying.

• Sampling Frame: A complete list of all population elements used as the basis for selection.

• Population Parameter: The true value of a variable across the entire population, such as the mean or proportion.

• Sample Statistic: The observed value of a variable calculated from the sample data used to estimate population parameters.

Estimating Parameters from Statistics

• For instance, considering a population of SFU students (37,000), if we select a random sample of 100 students to gauge approval of a gondola project (on a scale of 1 to 5), the average estimations of approval might vary across multiple samples, indicating sampling variability.

The Sampling Distribution

• The average of these sampling distributions trends toward approximating the true population average as sample sizes increase, showcasing the law of large numbers.

Estimating Sampling Error

• Standard Deviation (SD): Represents the spread of scores in the sample, providing insights into the variability among responses.

• Standard Error (SE): Indicates the standard deviation of the sampling distribution, which generally decreases as the sample size increases, reflecting increased precision in estimates.

Confidence Levels

• Various confidence intervals can be calculated from the standard error:

  • 68% Confidence: ±1 standard error

  • 95% Confidence: ±2 standard errors

  • 99.7% Confidence: ±3 standard errors

Random Assignment vs. Random Selection

Definitions

• Random Selection: Tied to the sampling process ensuring every member of the population has an equal chance to be included, crucial for extrapolating study results to the broader population.

• Random Assignment: Associated with the actual study design and is used to assign participants to different treatment conditions, enhancing internal validity by controlling for confounding variables.

Probability Sampling Methods

Key Requirements To conduct probability sampling, researchers must define the population, create a detailed sampling frame, and establish a predetermined sample size.Methods Include:

1. Simple Random: Each individual has an equal chance of being chosen through methods such as lottery or random number generation.

2. Systematic Random: Requires creating a random order and selecting individuals based on a defined interval (e.g., every 10th person).

3. Stratified Random: Ensures all relevant subgroups are represented by stratifying according to a key variable, leading to more nuanced insights.

4. Cluster Random: Effective in cases where individual sampling frames are impractical; involves sampling clusters or groups randomly rather than individuals.

Non-Probability Sampling Methods

Key Characteristics While simpler and often more practical, non-probability sampling methods carry limitations such as unknown probabilities of selection and limited generalizability.Methods Include:

1. Purposive Sampling: Selection based on research objectives or specific characteristics of subjects.

2. Convenience Sampling: Involves selecting readily available subjects, which may harm representativeness and lead to biases.

3. Quota Sampling: A non-probabilistic version of stratified sampling that sets fixed quotas to ensure that certain characteristics are represented.

4. Snowball Sampling: This method is particularly useful for accessing hard-to-reach populations by allowing existing participants to refer future subjects.

Next Steps

• Research Proposal Part 1 is due tonight at 11:59 PM.

• Prepare for the next lecture on Survey Research by reading Chapter 9 of the assigned textbook.