2801 Final Flashcards
Surveys
Are correlational research. Causality may be inferred. Some survey research makes predictions (predictor variables & criterion variables). Types include questionnaires, interviews, and self-reported diaries.
Designing Good Surveys
- Consider research question 2. Define Constructs 3. Review existing instruments 4. Write items for each construct 5. Get advice 6. Pilot test items 7. Analyze stats 8. Re-work items 9. Administer final survey
Defining Constructs
The most important stage in survey design.
Types of Questions
Demographic information, open-ended items (tend to be subjective, used more in interviews), close-ended items. Note: Interval > Ordinal data for demographics
Scales
Categorical (nominal), Continuous (interval / ratio), Ranked (ordinal), Scores can either be summative or cumulative.
Likert Scaling Model
Summative scaling method with ranked values –> anchors are susceptible to bias –> Even # scale has no neutral choice
Semantic Differential Scale Model
Summative model, measures feelings by scaling between 2 extremes. Only extreme anchors labeled.
Visual Analogue Scales
Summative, only extreme anchors labelled, line of fixed length used for scale. Commonly used for pain.
Guttman Scales
Cumulative or hierarchical. Often describe functional limitations of patients.
Order of question types (in a survey)
Non sensitive (interesting), demographic (non-interesting), sensitive info, end with easy questions
Delphi survey
A survey in which participants are health-care practicioners, or experts in the field –> Develop consensus around a specific issue –> Useful for establishing norms in clinical practice.
Emic approach
Qualitative, begins with indicators & tries to determine constructs that fit. Goal is to understand.
Etic approach
Quantitative, begins with formal constructs & tries to develop empirical indicators. Goal is to predict.
Variance Questions
Focus on differences & correlations. Focus on testing predetermined solutions (Hypothesis testing)
Process Questions
Focus on how things happen. Focus on understanding –> Hypothesis generating
Qualitative research
Concern is with discovery & description. Qualitative research in health studies SDoHaD
5 Qualitative research study types
Normative-Biographical study, Phenomenological Study, Ground Theory study, Ethnographic & Case study
Normative-Biographical Study
Researchers focus on the meaning an individual finds in his/her experience.
Phenomenological Study
Studies a phenomenon. Researchers focus on recall & recounting of marker events (key experiences that shape an indiv’s life)
Ground Theory Study
Focuses on finding relationships or various interpretations an indiv applies to his/her experiences –> Researchers develop constructs grounded in daily life experiences
Ethnographic Study
Focus on cultural patterns of behavior & meanings people use to organize & interpret experiences
Case Study
Analysis of a case to invoke broader interpretations of the meaning –> Structured: Problem, context, issues, lesson learned
4 Methods of Data Collection
Interviews, Observation, Content Analysis, Focus Groups
Interviews
One method of data collection: Structured (fixed questions) or unstructured (questions develop as interview progresses)
Observation
One method of data collection: Direct determination of “here and now” experiences. Either passive or participant.
Content Analysis
One method of data collection: In depth look at qualitative materials
Focus groups
One method of data collection: Investigators act as moderators, facilitating a discussion
Analysis Process
1) Data managing (data storage) 2) Reading / memoing (read, note, form initial ideas) 3) Describe data 4) Classifying 5) Interpreting 6) Representing & visualizing
Sampling
Evaluate saturation to determine sample efficiency –> When sufficient information exists to predict responses it is saturated. Saturation is difficult to quantify.
Tracy’s 8 “Big Tent” Criteria
1) Worthy Topic 2) Rich Rigor 3) Sincerity 4) Credibility 5) Resonance 6) Research provides Significant Contribution 7) Research is Ethical 8) Meaningful Coherence
Weakness of Qualitative research
Labor intensive, involves exp-based learning, lack of formal rigor in data collection & analysis, more difficult to establish credibility.
Variables
Independent vs dependent, predictor vs criterion, subject, continuous vs discrete
Levels of measurement
1) Nominal –> Categorical data w/no implicit ordering, unequal distance between points. 2) Ordinal –> categorical w/implicit ordering, unequal distance between points. 3) Interval –> Continuous (eq dist between points) and no meaningful 0. 4) Ratio –> Continuous, meaningful zero.
True Scores & Errors
Any deserved score (x) is comprised of 2 distinct components: (T) True score & (E) Error component. X = T + E. X - T = Measurement error.
Measurement Error
1) Systematic errors –> Predictable, reliable, thus more of a validity concern. 2) Random –> Occur due to chance. As Re’s decrease, T approaches X, measure becomes more reliable.
Validity
Construct validity (does measure represent what it’s supposed to), Internal Validity (are effects due slely to experimental conditions), External Validity (can results be applied to other settings or populations), Statistical conclusion validity (were appropriate methodological & statistical techniques applied)
Utility
Is the data precise & reliable at lowest cost (efficiency), can the method be applied (generality)
Measurement issues in the assessment of change
1) Lvl of measurement 2) Reliability 3) Stability 4) Linearity
Measurement issues in Lvl of measurement
Nominal scores can’t be subtracted, ordinal scores have unequal distances between points, Interval scores can be subtracted but amount of change can’t be computed –> Change is best measures by ratio measures
Measurement issues in Lvl of reliability
If measures are unreliable, your change score will contain mostly error. Suggested change scores only be used when reliability exceeds 0,5 but should exceed 0.7 in practical settings.
Measurement issues in Lvl of Stability
Important in situations where there may be substantial variability in performance.
Measurement issues in Lvl of Linearality
The shape of the relationship with time may effect measrements of change.
Evaluating Diagnostic Procedures
Sensitivity = test’s ability to obtain a “true positive” can be calculated by true positives over total positives, Specificity = test’s ability to obtain “true negative” can be calculated by true negatives over total negatives
Positive Predictive value (PV+)
Likelihood a person testing positive will have the disease Can be calculated by dividing true positives over total number of positive screens.
Negative Predictive Value (PV-)
Likelihood a person testing negative will be disease free. Can be calculated by dividing true negatives over total number of negative screens.
Cross-validation
Assessments of construct validity –> use a measure to predict group membership in a sample different from the one used to originally determine a cut off score
Criterion Referenced Tests
Interpreted relative to a standard that represents an acceptable level of performance
Norm-referenced tests
Interpreted relative to the performance of a peer group (established by testing a large group, and establishing cutoffs with the distribution)
Mean
Interval & Ratio only, it’s the average of the data
Median
Ordinal, interval or ratio, it’s the point that divides the data in half (n+1)/2
Mode
Nominal, ordinal, interval, or ratio, it;s the most frequently occurring vlaue
Central Tendency
If distribution is normal (Bell-shaped), mean, median & mode are all the same.
Dispersion
Range (ID highest & lowest values) & Standard deviation (more accurate & detailed)(shows relation that indiv scores have to the mean sample)
SD formula
draw
SD (Computational formula)
draw
Normal Distribution
“bell curve” –> Mean, median & mode is 0, standard deviation is 1. 68% w/1SD, 95% w/2SD, 99% w/3SD
Standard scores
Easiest way to compare scores on a common scale is using z or T score
Z-Score
Is a standard measure of the distance between a single point in the data & the overall mean for that variable.
Z-score formula
Draw
Z-distribution
ranges from - infinity to + infinity, mean of 0 and STD of 1
T-score
Is a standard distribution with a mean of 50 & std of 10 and no negative values.
T-score formula
T = 10z + 50
Histogram
Compares multiple measurements of the same variable
Bar graph
Compares multiple variables
Stem & leaf
A vertical histogram, shows raw data & gives a rough idea of dispersion.
Frequency polygen
Similar to a histogram, useful in summarizing interval-level data
Line graphs
Often used to convey temporal information
Box Plots
IQR = QU - QL, outliers fall outside boundary set by median (+/- 1.5 x IQR), extreme outliers fall outside boundary set by median +/- 3.0 x IQR)
5 Number summary (Box plots)
Min value, lower quartile, median, upper quartile, max value
Notched box plots
Like box plot but adds 95% CI. Median > mean = negative skew, median < mean = positive skew. IQR = middle 50% of data.
Hypothesis
Is our best guess. Research / alternate hypothesis (best guess) vs Null hypothesis (nothing happened). We always test against null hypothesis. Can never accept null hypothesis only able to reject or fail to reject.
Directional Hypothesis
A specific result being tested in a direction from the mean (upper tail or lower tail test)
Non-directional hypothesis
One is comparing change that may occur in either tail of the distribution (two-tailed) –> 2 critical values & therefore two rejection areas for null hypothesis
Alpha
Probability of incorrectly concluding that there is an effect (T1 Error)
Power
Ability to determine true relationships that exist within the data.
Determining power
Power is equal to 1 - beta. Where beta is a type 2 error.
T2 Error
When Ho is false but you fail to reject Ho.
Estimating Alpha
Know Z-score formula, apply to Z table and subtract z from 1 to determine alpha
Central limit theorem
As sample size increases, approximation of normality in the sampling distribution improves, aka the normal convergence theorem.