Research Methods PT. 3 Flashcards
Population
the entire set of people and things that you are interested in
- all first year students at unc
Sample
smaller set of people that’s taken from the population
-sample of sophomores from UNC
Census
sample every member of a population
biased sample
not all members of the pop have an equal probability of being included in the study
representative sample
all members of the population have an equal probability/chance of being included in the study sample of first year students from UNC if interested in the population of UNC
ONLY UNBIASED SAMPLES ALLOW US TO MAKE INFERENCES ABOUT THE POPULATION OF INTEREST
When would a sample be biased?
-A researcher’s sample might contain too much from unusual people
Ex: reaching to the bottom of a chip bag and only eating the crumbs
-A sample might only include data from one kind of people
Ex: poll for men and women but they only only sampled men
Ways that a sample could be biased?
Sampling only those who are easy to contact-convenience sampling
May not reflect the entire population,
Psychology, davie hall, get people who are seeing that are coming to davie hall
Self-selection: sampling only those who volunteer
Can cause serious problems for external validity
Because people who are convenient or more willing might have different opinions from those who are less handy and less willing.
Probability sampling
Every member of the pop has an equal chance of being in the study
Simple random sampling: every members name in a pop of interest, in a pool and randomly select a predetermined number of numbers
Systematic sampling:
Roll 2 dice, start at 5th person and then go 3 people over
Cluster sampling
Clusters of participations within a pop of interest and a cluster is randomly selected, all members from cluster are selected
Multistage
Clusters but there is a random number of people from that clusters
Step 1: cluster is chosen, step 2: people form cluster are randomly selected
Stratified Random Sampling
Interest in specific demographic categories
Race gender sexuality and then you randomly select via those categories
Oversampling:
Over Represent one or more groups. More than the actual percentage of people in that pop
Random sampling
Creating a sample using a sort of random method
Increases external validity
random assignment
used only in experimental designs to assign participants to groups at random
Increases internal validity
Correlation coefficient
r
direction tells us that an association is positive negative or no association or zero association
if one variable is categorical
look at the mean instead of the correlation coefficient
bar graph, not associating them with meaningful numbers
Statistical Validity
how strong is the relationship?
- effect size, strong 0.5, 0.3 moderate, and 0.1 small or weak
0.3 fairly powerful
-0.4 and above is strong, 0.2 moderate, .10 small or weak, .05 very small
small effect size
can compound over many observations
confidence interval
contain true population correlation, CIs do not contain zero, if they do the finding is not statistically significant
bigger range
less precise or stable
Outliers
One the ways to know if you have outliers is looking at scatter plots
One or more extreme scores that lie far away from your data
One single outlier can have a sig impact on your correlation coefficient
Depending on where it is it can make a stronger correlation weak or weaker correlation
Outlier is more influential with smaller sample size
If you take the outlier out and don’t include it in the analysis it can have a very large effect that changes it significantly
restriction of range
when there is not a full range of scores in the association-can maybe make the correlation look smaller than it is does
Restriction of range – when there is not a full range of scores on one of the variables in the association – can make the correlation look smaller than it is. Can use stat techniques to correct for the restriction of range or recruit more participants to add data points to study
concurrent validity
Amount of agreements between two different assessments that measure the same construct
Going to compare a measure we have to a gold standard measure- when we are thinning about the same construct with two:
Ex: intelligence, assessment of intelligence
Stanford Binet Intelligence Scale
Wisch wechsler intelligence scale for children
discriminant validity
extended to which attest we have is no related to other tests that measure different constructs
Two test measuring two different constructs, don’t expect them, the results should not be similar
Ex; of happiness ex of sadness, your scores should be noticeably different or not related they are definitely measuring two constructs
convergent validity
How the measure of one construct aligns with a measure of the same or related construct
concurrent validity vs convergent validity
Concurrent validity compares a measure to a gold standard measure but convergent validity compares two measures of similar/ related constructs rather than having a gold standard measure, type of concurrent validity
Curvilinear Association
correlation coefficient is zero, no association between data points and the data points are not a linear format to determine a distinct correlation
In this particular relationship as people’s ages increases
No relationship, curvilinear relationship, the dat
U RELATIONSHIP or inverted U
third variable
third variable has to correlate to both original variables