Social Science Quantitative methods(03) - Research Design

 

Research Design

1. True Experiments

Randomized control trials/RCT

Maximize internal validity

  • 3 essential ingredients for true experiments:

    • Manipulation

      • cause under control
      • eliminate temporal precedence: cause precedes effect
    • Comparison

      • control group
      • ensures effect does not occur naturally: maturation
    • Random assignment

      • ensures there are no other explanations for the effect
      • Replication
        • is required to ensure randomization works

2. Factorial Design 析因设计

  • several independent variables–factors
    • investigated simultaneously
  • one dependent variable

3. Repeated Measures

  1. Between Factor 组间因素

    • the dependent variable measured after being exposed to the independent variable.
      • gender
  2. Within Factor 组内因素

    • the dependent variable measured after being exposed to each level of the independent variable.

      • different level of dosages to one patient
    • Repeated measures design 重复测量设计

      • measured repeatedly
      • Short period
    • at least one within factor and one or more between factors
  3. Longitudinal design 纵向设计

    • measured repeatedly

    • long period of time

    • rare

    • mostly correlation studies

    • sometimes (quasi-) experimental studies

4. Research Control

  1. Manipulation

    • control over independent variable

      • value/level determined by researcher
    • control over external variables

      • rule out alternative explanations
    • terms: level;condition;group

    • experimental variable

      • independent variable that can be fully controlled
    • individual difference variables

      • intrinsic property of participant that cannot be controlled
      • age/gender
    • Seemingly non-manipulable variables that can be manipulated
      • a manipulation check
      • measuring after the dependent effect variables
  2. Control of variables of disinterest

    ideal: only difference: independent variables
    • ceteris paribus: all other things are equal

    • much harder in social science to control

    • Control variables

    • solution:

      • Randomization随机化

        eliminates all systematic differences between participants in different conditions on average

        • Randomization failure: unequal distribution possible

        • solution:

          1. Randomization check: measure relevant variables to check whether randomization worked and the conditions are the same.

          2. Restricted randomization(限制性随机)

            guarantee randomization over selected variables

        • blocking(区组): guarantees equal group sizes

          • pair participants in blocks of two
          • flip a coin to determine where the first participant goes
          • the second… participant is automatically assigned the other condition
        • stratified restricted randomization(分层限制性随机): equal distribution on participant characteristic

          • boys/girls
          • sex
          • age
        • Limits:

          • >2,3 variables–>too complicated
          • too many possibly relevant variables
          • Randomization with repeated measures
    • It is necessary!

    • avoiding maturation

      • counterbalancing: order randomized抵消平衡法
        • Within subject designs: the order in which subjects are exposed to the conditions should be randomized.
        • Subjects are assigned to one out of all possible orderings of the conditions possibly using blocking to ensure that no one ordering is over represented.列出所有可能的实验条件序列 并将受试者分配到每一组 尽量使用区组来确保各组人数相当
  3. Matching

    • Random assignment impossible?
      • threat of selection to internal validity
      • pragmatic/ethical reasons
      • independent variable–>individual differences variable
    • Matching on relevant background variables
    • Threat: undermatching
      • measured with some error
      • related to the variables of interest
      • Due to regression to the mean!(均值回归)
        • 选择的对象没有代表性导致重复试验时向均值回归。
      • detrimental effect!

5. Lab vs Field

  • Lab

    • often experimental
    • maximizes internal validity
    • low ecological validity? mundane realism?
      • not similar to natural setting
      • Low ecological validity doesn’t imply low construct validity or low external validity
    • experimental realism
      • experimental translation of natural setting
  • Field

    • often non-experimental
    • maximize external validity
    • uncontrolled environment
    • often observing natural behavior

6. Experimental Designs

  1. Two-group Design双组设计

    • Experimental group: the hypothesized cause is present

    • Control group: the cause is absent

    • Two-group pretest/post-test design

      • avoid maturation
      • compare the size of the increase or decrease in scores
    • pretest

      • sensitize 被试敏感化
      • practice effect 练习效应
        • higher score in post-test
      • alert participants to the purpose of the study
  2. Solomon four-group design 所罗门四组设计

    • two group design+ two group pre/post-test design

      • the experimental/control condition are run twice
        • one with a pretest
        • one without a pretest
      • avoid practice effect
  3. Repeated measure design with one within subject factor 单一主体因素重复测量设计

    • all participants are exposed to all levels of the independent variable. They experience all conditions.

7. Quasi-experimental designs

  • non random assignment

    • random assignment can be: impractical/unethical
  • sometimes no manipulation, or comparison

    • individual differences variable: non-manipulable
      • sex
      • disease
  • investigates causal relation

  • natural event/natural experiment

  • values selected, not manipulated

    • like correlational studies
    • difference: 准实验研究检验的是因果假说 且会尽可能地极力控制额外变量.
  • Types of designs

    1. Static groups comparison

      Static groups comparison

      • similar to the two group experimental design
      • a post-test after exposure to the independent variable
      • Difference: Non random assignment
        • internal selection threat
        • Comparison of groups lessens the threats of maturation and history
    2. Pre/post-test Non-equivalent control group 不等组前测/后测设计

      Pre/post-test Non-equivalent control group

      • firmly disambiguate the causal order
      • assess the size of any maturation effects
      • not always possible to use in control group
    3. One group pre/post-test 单组前测-后测设计

    4. Interrupted time-series design 中断时间序列设计

      interrupted time-series design

      • more observations/ tests before and after the treatment

      • avoid maturation

      • still history threat

        • solution:

          • replicated interrupted time-series design

            • control group
            • pre/post-test non-equivalent control group with more tests :smile:
          • redo the experiment at a different time

            redo the experiement

8 Correlational design相关性研究

  • no manipulation of independent variable

  • or no independent variable identified

  • no causal direction specified

  • don’t select/manipulate just measure

  • types:

    correlational design

    1. Cross sectional design 横断面实验设计

      • one large group is considered at one specific point of time
      • a fairly large number of properties are measured at once
      • aim: to investigate the association between variables in a way that accurately describes a larger population
      • survey study or survey design
    2. time-series design 时间序列设计

      • one person is measured at several points in time, usually with many measurement moments in quick succession and in fixed intervals.

        • longitudinal design vs. time-series vs. repeated measures
          • longitudinal
            • general term for measuring repeatedly over long period of time
          • time-series
            • correlational: 1 person measured over time
            • quasi-experimental: people in conditions measured repeatedly
            • experimental: N=1 study: 1 person measured over time in all conditions
          • repeated measures
            • experimental: persons measured over time in all conditions where
            • one or more independent variables are manipulated
    3. Panel design 群组实验设计

      • also call time-series cross-sectional design(TSCS)
      • the same group of people is measured at several points in time
        • cohort design 队列实验设计
          • a special case of panel design
          • participants are all same age or started a new school or a job at the same time

9 Other designs

  • case study

    • one person/group(as as whole)
    • quantitative: n=1, time-series
    • qualitative: generating hypotheses more than testing
    • negative analysis
      • contradictory evidence
      • counter-evidence against a hypothesis
  • evaluation research 评价研究

    • applied research, non -experimetnal
    • broader scope
    • effect of policy/program
    • summative: outcome of a program
    • formative:process assessing how a program works
    • sociology, communication sciences, political sciences
  • Intervention study 干预研究

    • experimental
    • effect of treatment on individuals
    • developmental/clinical psychology
  • Validation study 有效性研究

    • assessing the quality of an measurement instrument
      • surveys
      • questionnaires
      • tests
    • statistical analysis
    • psychometrics, sociometrics