L7 - Total Survey Error

Data Generation Techniques

• Organic data is generated incidentally without specific intention, often a byproduct of regular processes.
• Purposive data is collected with a specific research question in mind, structured to meet the study's objectives.
• Designed data is generated via controlled methodologies like surveys and experiments, aimed at describing social reality, studying causal relationships, or generalizing to the target population.

Total Survey Error Types

• Coverage Error: Occurs when certain population members are excluded from sampling, leading to skewed results if excluded individuals differ from those sampled.
• Sampling Error: Arises from surveying a subset of the population; increases with smaller and non-random samples, contributing to uncertainty requiring statistical quantification.
• Nonresponse Error: Results from selected individuals refusing participation (unit-nonresponse) or specific questions (item-nonresponse), potentially causing bias if respondents differ systematically from non-respondents.
• Adjustment Error: Occurs from improper weighting or handling of data that leads to underrepresentation of certain groups; can be corrected through methods like “weighting” or imputation.
• Measurement Error: Arises from faulty responses due to survey design issues or respondent behavior, impacting data integrity.
• Processing Error: Involves errors during data collection and processing stages, including data entry and handling mistakes.

Correlation and Hypothesis Testing

• Correlation measures the strength and direction of relationships between two variables, with the Pearson correlation coefficient ranging from -1 to +1.
• The sampling distribution reflects variability in correlation coefficients (r) across multiple random samples from the same population.
• Population correlation is represented by ρ; a larger sample size leads to a smaller standard error.
• In the absence of a relationship (ρ=0), sample distributions will be centered at zero, with both positive and negative correlation values observed.

Statistical Hypotheses

• Null Hypothesis (H0): Assumes no correlation exists in the population (ρ=0).
• Research Hypothesis (Ha/H1): Proposes a positive correlation exists (ρ>0), linking kindness and life satisfaction.
• A large p-value indicates strong support for the null hypothesis; a small p-value suggests little support for it.
• Chance refers to the probability of observing a sample value as extreme as the one obtained, evaluated using mathematical models and represented by areas under the probability distribution curve.