Measuring
Statistical Reasoning - Measuring
Measurement Basics
For any variable of interest, it is essential to decide how to measure them.
The result of a measurement produces a numerical variable that can have different values.
Problem Identified:
Some variables can be challenging to measure due to their abstract or unobservable nature, examples include:
Personality
Work ethics
IQ
Ambition
Certain variables represent abstract concepts, such as happiness, satisfaction, and motivation, which are difficult to measure directly.
Solutions to Measurement Challenges
A solution to measuring abstract variables is to utilize an instrument, which is defined as a variable that can reasonably measure the abstract concept.
Example of Measurement Instrument:
Ambition can be measured through years of education, which serves as a proxy for ambition (an abstract and hard-to-measure variable).
Researchers commonly rely on proxies, surveys, or scales to quantify abstract concepts as accurately as possible.
Valid and Invalid Measurements
A variable is considered a valid measure of a property if it is relevant or appropriate as a representation of that property.
Valid Example:
Measuring length with a tape measure is valid.
Invalid Example:
Measuring a student's readiness for college using their height is invalid.
The validity of an instrument used to measure a variable relies on how convincingly the researcher justifies that it accurately measures the intended concept.
Examples of Validity in Measurements
Bureau of Labor Statistics (BLS) Unemployment Rate:
This measure is considered valid, despite potential variations with changes in official definitions.
Generally, a rate (a fraction, proportion, or percentage), which occurs in the interval [0, 1], serves as a more valid measure than a simple count of occurrences, facilitating easier comparisons.
SAT Scores as a Measurement Tool
The validity of the SAT as a measure for readiness for college is complex:
"Readiness for college academic work" is a vague concept that likely involves multiple factors.
Opinions may vary on whether SAT scores accurately reflect this concept.
Instead of focusing on the validity of the SAT scores outright, the question becomes:
Do SAT scores help predict students' success in college?
A variable (e.g., SAT score) is said to have predictive validity if it can be adequately used to predict another variable (e.g., student success in college).
A crucial goal of most data analysis is to build models that can make predictions.
Including a variable X in a model implies that X holds some predictive power for another variable, Y.
Measurement Accuracy and Errors
Measurements of abstract concepts often entail errors that contribute to bias and variability.
To enhance accuracy, multiple measures may assess the same concept (e.g., different survey questions about personality).
Predictive Validity:
Recognized as the clearest and most valuable type of validity from a statistical perspective.
It is essential to recognize that measurement errors exist, described mathematically as:
\text{Measured Value} = \text{True Value} + \text{Bias} + \text{Random Error}
Reducing Bias in Measurements
A measurement process exhibits bias if it systematically overstates or understates the true value of the property being measured.
Challenges in Reducing Bias:
There is no straightforward method to reduce measurement bias.
Bias is influenced by the quality of the measurement instrument.
To mitigate bias in measuring variables:
Acquire a superior measurement instrument.
Reducing bias is typically more complex than reducing variability.
Simply increasing sample size does not alleviate bias since it originates from flaws in design or measurement, rather than random variation.
Reducing Variance in Measurements
A measurement process has random error if repeated measurements on the same individual yield different results.
To evaluate whether random error is minimal, compute the variance.
A reliable measurement process exhibits a small variance.
Strategies to reduce variance (and thereby enhance reliability) include:
Taking more samples.
Reliability:
Similar to the concept of repeated sampling, reliability assesses the consistency or variability of measured values.
A reliable measurement produces similar results under consistent conditions.