Chapter 5, 8, 9 & 4 Flashcards
Social Research - Approaches and Fundamentals
Purpose of research
Collecting info reagrding target population
Steps for sampling
- Defining target population
- Obtaining sampling frame
- Deciding sampling design
Coverage error
Differences between the target population and the sampling frame, when the sampling frame does not include all members of the population. Ex. telephone interview omits people without a phone
Two types of coverage error
- Undercoverage: Belong to the population, but are not represented in the sampling frame (most problematic!!)
- Overcoverage: are not part of the target population, but are listed in the sampling frame
Sampling designs
Part of research plan that indicates how cases are being selected for observation
Sampling designs divided in 2 classes
- Probability sampling (4 ways): all cases are randomly selected
- Non probibility sampling: cases are non-randomly selected
Random selection
Each element in a set has an equal chance of being selected. If not, it is called biased
Simple random sample
Equal chance in random selection
Sampling error/random error
Differences between a population characteristic (parameter) and the sampling estimate (statistic) of that characteristic
Systematic error
Simple random sample only leads to representative info when the coverage error is small, if not, it is called systematic error
Confidence interval
Tells us that interval estimates obtained with the procedure are likely to contain the population parameter 95% of the time
Stratified random sampling
Stratification of sampling frame in group of elements that share the same characteristics (strata). Guarentees that even smaller strata will be represented in sample
Disproportionate stratified random sampling
Variation of random sample. Unequal chance of selection depending on how large stratum is
Cluster sampling
The population is broken down into groups of cases, called clusters
Two ways to sample
- Single staged cluster sampling –> sampling with cluster level. Primary sampling units.
- Multistage cluster sampling (nested sampling). Secondary sampling units.
Systematic selection
Consists of selecting every case from a complete list or file of the population, starting with a randomly chosen case from the first K cases on the list. 2 Requirements (Sampling interval & a random start)
Sampling interval
The ratio of number of cases in the population to the desires sample size
A random start
Refers to the process of randomly selection of the initial case between 1 and K.
Characteristics of survey research
- Large scale probability sampling
- Pre-constructed questionnaire/interview protocol
- Coded answers & statistical analysis
Descriptive surveys
Describe the distribution within a population of certain characteristics attitudes or experiences
Explanatory surveys
Beyond description to investigate relationship between two or more variables and attempts to explain these
Research design
Overall structure of a study
Cross-sectional design
To represent a particular target population are gathered at one point at a time. (Single measurement + single time sampling)
Two variations of Cross-sectional survey
- Contextual design: sample enough cases within particular groups to describe accurately certain characteristics.
- Social network designs: focus on the relationship or connections among social actors.
Longitudinal designs
Same questions are asked at two or more points in time
Two types of longitudinal designs
- Trend study: Same questionnaire, same target population, different sample. See changes on group level.
- Panel studies: same questionaire, same target population, one sample with repeated measure. See changes on individual level
Time in sample bias
Participation becomes routinezed to the point that respondents repeat what they have said before
Panel conditioning
Due to repeates measures, stimulates attitude changes, so that respondents attitudes are no longer representative of the larger population
Four types of errors that threaten accuracy of survey results
- Coverage error
- Sampling error
- Non response error
- Measurement error
Measurement error
Inaccurate responses associated with the respondent
Types of survey research
- Classification (survey modes) –> interviews, questionnaire
- Optimal sampling design (from realistic perspective) –> Face tot face, mail/telephone
Types of response scales
Dichotomy, Rating questions, Picking/ranking scales
Dichotomy
Most ‘easy’ one, just two options. yes/no
Rating questions (labels/numbers)
- Likert scale (5 point scale)
- Thermometer question ( only end labelling of response categories, in between using values)
- Semantic differential (two opposites opinions)
Picking/Ranking scale
- Ranking (full/partial)
- Picking (without ranking)
Primary effect
More likely to choose an earlier mentioned option
Recency effect
When you are more likely to choose the last given options.
Conceptualization
Describe and defining of concepts that one aims at measuring/observation
Two important aspects to the measurement process
- A single category (ex. male, b-grade student)
- Several values (ex. gender, exam results –> multiple items)
Operationalizing
Making the concept measurable
Index
Multiple item indicator. Reduce data generated by multiple indicators into a single numer or scale score –> If indicators are combined without a formal ‘test’ of its scale characteristics
Reliability
The precision & consistency in measurement
Validity
Congruence between an opearional defination and the concept it is purported to measure
Measurement error
Points of departure: Classical test-theory:
X= T + Es+ Er
Systematic measurement error
Validity problem
Random measurement error
Reliability error
Test-retest reliability
Testing/measuring the same persons/units on two seperate occasions with the same measurement. correlation 1 is perfect, should be higher than 0.8
Split half and internal consistency
Multiple items measuring one single concept; one measurement
Split half method
Divide items of a scale into halves. There will be a correlation between these two subsets
Internal consistency
Examines the relationships among all the items (=Cronbachs alpha)
Face validity
Personal judgement
Criterion related validity
Applies to measuring instruments that have been developed for some practical purpose
Construct validity
Emphasizes the meaning of the responses to one’s measuring instrument
4 types of construct validity
- Correlation with related variables
- Differences between known groups
- Consistency accross indicator and different methods of measurement
- Correlations with unrelated variables
Heterogeneous target (language)
Use common language / keep it simple
Homogeneous target (language)
Use language of your targetgroup (ex. doctors)