By Sarah Boslaugh; O'Reilly Media, Nov. 2012
ISBN 9781449316822
- Types of data
- Nominal - unordered, non-arithmetic
- Ordinal - ordered, non-arithmetic
- Intervals - ordered, equal intervals between measurement points, non-arithmetic
- Ratio - ordered, equal intervals, has a natural zero point, meaningfully divisible
- Continuous - can take any value within a range (floats)
- Discrete - can only take specific values (counts, integers)
- Classical measurement theory conceives of any measurement or observed score as consisting of two parts: true score T and error E
- X (observed measurement) = T + E
- Random error is everywhere, and can be dealt with or assumed to cancel out
- Systematic error:
- has an observable pattern
- is not due to chance
- often has a cause or causes that can be identified and remedied
- Methods of measurement have to be evaluated for reliability and validity
- Reliability - how consistent and / or repeatable measurements are
- Multiple-occasions reliability, or test-retest reliability, is how similarly a test or scale performs over repeated administration. Can be a measure of 'temporal stability.'
- Multiple-forms reliability, or parallel-forms reliability, is how similarly different versions of a test or questionnaire perform in measuring the same entity
- Internal consistency reliability is how well the items that make up an instrument reflect the same construct (how much they measure the same thing).
- Validity - how well a test or rating scale measures what it is supposed to measure
- Content validity - how well the process of measurement reflects the important content of the domain of interest
- Face validity - whether it appears to a typical person to be a fair assessment of the qualities under study. Important for establishing credibility.
- Concurrent validity - how well inferences drawn from a measurement can be used to predict some other behavior or performance that is measured at approximately the same time
- Predictive validity - how well you can draw inferences about the future
- Triangulation - using information from multiple sources to arrive at some accurate or more accurate output, or to adjust for error in one source
- Measurement bias
- Can enter studies in two primary ways:
- during selection and retention of subjects
- in the way information is collected about subjects
- Selection bias - when some potential subjects are more likely than others to be selected for the study sample
- Volunteer bias - people who volunteer for things aren't usually representative of the population as a whole
- Nonresponse bias - people who decline to take part represent a portion of the population, but are not in the results
- Informative censoring can create bias in any longitudinal study. The worst of that is when study participants drop out not at random but for reasons related to the study's purpose.
- Information bias - enters through the data collection methodology
- Interviewer bias - introduced by the knowledge or attitudes of the researcher
- Recall bias - people with a life experience such as suffering from a serious disease are more likely to remember events that they believe are related to that experience
- Detection bias - certain characteristics are more likely to be detected or reported in some people than in others.
- Social desirability bias - people want to present themselves in a favorable light
- Can enter studies in two primary ways:
- Trials - experiments, observations, some event whose outcome is unknown
- Sample space - set S of all possible outcomes of a trial
- Event - E, the outcome of a trial, can consist of one outcome or a set of outcomes
- Union - compound event that occurs if one or more of the events occur. EuF means "either E or F or both E and F"
- Intersection - compound event that occurs if all the simple events occur. EnF means "both E and F"
- Complement -