How can we help?

You can also find more resources in our Help Center.

60 terms

Biostats/Epi Final

STUDY
PLAY
Face validity
-Not recognized by all theorists
-"on the face' of it approach
Content validity
Answers the question: How much a measure covers the meanings within a concept?
Construct Validity
-Logical relationships among variables
-Offers evidence of whether your measure does or does not address the construct
Criterion Validity
-Degree to which a measure relates to some exterior criterion
-Sometimes known as predictive validity
-I.e. PCAT scores
Concurrent Validity
-Degree to which a measure relates to other similarly validated tests
-E.g. The Test of Functional Health Literacy (TOFHLA) was compared to REALM
External Validity
-Generalizability
-Test results across different populations
-Applicability of results
Generalizability theory
-Introduced by Cronbach and colleagues
-Bridging reliability and validity
-Use of ANOVAs
-Item difficulty vs. Person by item interaction
Survival Analysis
Studies that involve prospective analysis i.e. following individuals from one point in time to outcome (usually death)
-Can be used for any end-point/outcome: e.g. smoking cessation, tumor recurrence etc.
-End-point and follow-up time must be documented
Censoring
-value of a measurement or observation is only partially known
-don't have exact event time
Reasons for having censored data
-Subject does not experience event before study ended
-Loss of follow-up
-Withdrawal from study
Life table analysis
-Can be thought of as enhanced frequency distribution table
-Divide observation interval into smaller periods of time to estimate the outcome during each period
Kaplan-Meier Curves
-Similar to life tables
-Uses exact times that events occurred instead of intervals of follow-up
-P(e) = Number of events at that time divided by Number at risk at that point in time
Cox proportional hazards model
-Uses proportional hazards
-Key to interpretation is use of odds ratios or in some cases, hazard ratios
Problems with open ended questions in surveys
-Need to be coded for analysis
-Coding is interpretation and may lead to researcher -bias
-Irrelevant answers may be a problem
Important considerations for closed ended questions in surveys
-Important that categories be mutually exclusive
-For multiple responses to a problem - coding difficulties arise
Key factors of question construction in surveys
I. Make items clear
II. Avoid double barreled questions:
III. Respondents must be competent to answer
IV. Respondents must be willing to answer
V. Questions should be relevant
VI. Short items are preferable
VII. Avoid negative items:
IX. Avoid biased items and terms
Thurstone Scaling
-One of the first scaling theorists
-Method of equal appearing intervals, successive intervals and paired contrasts
Likert Scaling
-Develop the focus: e.g. generate statement about specific attitudes that people may have toward
-Item generation: 1-5 or 1-7, Agree to Disagree scale
-Rate items: group of judges rate items from strongly unfavorable to strongly favorable to concept
-Select items: throw out unfavorable items (may use item correlation - easily done in SPSS)
-Administer items: 5 point likert scale
Guttman Scale
-Cumulative scaling
-Develop focus e.g. attitudes of patients toward immigrant physicians
-Develop items: e.g. I would feel comfortable discussing my health problems with an immigrant physician
-Rate items
-Develop cumulative scale: Construct matrix
Reliability
reproducibility, repeatability or consistency
Inter-rater reliability
-Aka Inter-observer reliability
-To check for consistency from one observer to another
For measuring reliability a general rule of thumb is ______indicates good level of agreement
Kappa >0.6
Test-retest reliability
-Administer test on two different occasions
-Time is important: shorter time gap - higher correlation and vice versa
Parallel forms reliability
-Create two parallel forms
-Correlation indicates level of reliability
-Limitation: need a large set of items
Statistical Decision Theory (SDT)
This answers the question: How often are we making good decisions? How often are we making bad decisions?
Type I Error or alpha
When you reject the Null Hypothesis but it is actually true Probability you say BS when H0 is really true
Type II error or beta
Accept the Null Hypothesis when its False (Say nothing is going on when something really is going on)
Power
1 minus beta
Basic Factors that Influence Power
A) Size of the difference between means
B) Significance Levels
C) Sample Size
D) Variance or standard deviation
E) Experimental Design
Why do large samples increase power?
1) You decrease the "appropriate sd" sdiff, sm, sm, etc
2) You reduce the effects of outliers and sampling error (central limit theorem) less likely to get weird sample with bigger n
How do you decrease variance in your sample
1) Increase Sample Size
2) Decrease Error in your Research Design
3) Use a Homogeneous sample
Small effect size
d = .20
Medium effect size
d = .50
Large effect size
d = .80
Kappa
Po - Pe / (1-Pe)
Total sum of squares
measures the total scatter of scores around the grand mean. The grand mean is the mean of all the subject's scores regardless of the group to which they belong.
Between groups sum of squares
measures the total scatter of the group means with respect to the grand mean.
Within Group sum of squares
measures the scatter of scores within each group with respect to the mean of that particular group. .
df in Anova between groups
Number of groups - 1
df in Anova within groups
Total number of people - total number of groups
df Total in Anova
Total number of people - 1
F is calculated this way
MSBW/MSWG
Post hoc analysis
-Test all pairs of means to see if they are diff from each other
-Done only after you find there is a significant difference in ANOVA
You need this information to find the q value in post hoc analysis
df of between groups and number of means in ANOVA
Two-Way ANOVA
-Uses F statistics to compare the differences among means for 2 independent variables (categorical)
-Interested in the effects of multiple independent variables on a dependent variable
Assumptions for Two-Way ANOVA
1) Interval or Ratio Data
2) Subject must be randomly assigned to cells
3) Populations from which samples are drawn must be n normally distributed
4) Populations from which samples are drawn must have equal variance
df when setting up t- in a correlation
df = N-2
Coefficient of Determination = r2
-tells us how much of the variance in one variable is accounted for by the variance on another variable.
-What do changes in X tell us about the value of Y. Essentially the "effect size" for the relationship between 2 variables
Phi Coefficient
Use this stat if X is categorical Y is categorical for correlation
Point Biserial Correlation
Use this stat if X is categorical Y is continuous
Spearman's Rank order Correlation Coefficient
Use this stat if X is ordinal and Y is ordinal for correlation
Homoscedasticity
This assumption means that the
variance around the regression
line is the same for all values of
the predictor variable (X).
Two approaches to research logic
Deduction - scientific method
Induction - start from observations
Types of qualitative research
Case Study
Grounded Theory
Phenomology
Ethnography
Historical
Case Study
Can be used to study a complex issue or extend experience and knowledge about a particular process
Emphasis is on detailed contextual analysis
Has often been dismissed by critics as a exploratory tool
Grounded Theory
-Theory developed inductively from data
-Case perspective approach
-Cases that are similar on many variables are compared to reveal necessary causes
-Methods: May include reading large volumes of textual data, observed behavior etc.
Phenomology
Field of philosophy
Study of structures, experience or consciousness
In other words, the study of phenomena
Studies in the first person to describe a wide range of experiences/emotions
Ethnography
Field of Sociology/ Cultural anthropology
Close field observation of socio-cultural phenomena
Ethnographer: one who conducts ethnographic studies
Works in a community, selects informants to interview, interviews informants multiple times etc.
Historical research
Identifying problem of interest that is historical
Conducting data collection and analysis based on interviews, examination of historical documents, biases in previous information collected
Quantitative facts if any
More than one point of view
Qualitative Research Methods
Participant Observation:
Direct Observation:
Unstructured Interviewing