Lecture 8 Flashcards
Hypotheses must be logical -> steps:
- Develop a hypothesis about a relationship
- Collect data
- Construct a statistic for that relationship ( t, F, chi-squared)
- Calculate the value of that statistic from the data,
- check the table
- If your result is unlikely to be caused by chance, you declare significance
Significant relationship
If there is enough chance that the value falls in the outer two tails of the distribution
Calculating the t-value using OLS
Y is your dependent variable, x independent with an error term
OLS
Ordinary least squared with particular assumptions least is minimal
Primary data
Collected by the researcher (survey, interview, experiment)
Secondary data
Existing data collected by another person (save time and money and non-obrustive, but not collected for the purpose of your study) also archival data
Dichotomous question
has two choices (a coded dummy variable)
nominal question
(non-ordered) multiple options, but no inherent order
ordinal question
ordered, multiple questions with an inherent hierarchy
Likert scale
an interval level ordinal question (5 to 7 points)
Question content what to ask and what is wrong
- Is the question necessary and useful?
- No double/mulitple-barreled questions (two questions in one)
- Do respondents have the needed knowledge?
- Can you make the question more specific or factual?
- How do you deal with sensitive questions?
- Take care of your wording ( not loaded, objective, etc.)
Common method bias
collecting data with a single method or source.
Solution: obtain x and y from different methods or sources, times/spaces
Self-serving bias
individual tendency to refer to internal attribution for their success and external attribution for their failure
Non-response bias
sustematic differences between respondents and non-respondents (solution: show that there are no differences)
- compare early and late respondents if you have no non-respondents data
Back translation
back translation is good for when things are lost in translation ( ask someone else to translate back and see if the message is the same)
before administering the survey
before administering the survey are variables missing? Are answers coded?
Probability sampling
form of random selection, representative or target population
Simple random sampling
select n cases out of N with an equal chance
- Systematic random sampling
- stratified random sampling
- cluster random sampling
Random sampling avoids sampling error
Nonprobability sampling
Use latter two if random is not available, It does not involve random selection
Three types: convenience, purposive, quota
Convenience
select conveniently available cases
purposive
attempt to obtain a sample representative of target population (if you have an idea how the typical case should look like)
quota
include observations until you achieve a specific number of cases for each stratum of the population
Statistical power: typer 1 error
Concluding hypothesis is supported when it is not true.
More stringent criterion alpha means the smaller probability
Statistical power: typer 2 error
Failing to find support for the hypothesis when it is true