Combining quantitative and qualitative methods – Advanced Research Methods – Joyce Rommens
Lecture 6: Synthesis: quantitative and qualitative methods
Quantitative part
Objectives segment 1:
- Explain the three different goals for quantitative research (associations)
- Explain how the strategy for descriptive research differs from the strategy for causal inference
1960-2010: methodological development focused on statistical methods
- Development of new techniques
- Improvements in computers and software
- Standardized tests, ‘objectivity’
- Helpful and harmful ® we can do more and better research, but people stopped thinking
themselves and rely on the software
® Associations can be seen in the data, but it is nothing more. What do they mean is up to the
researcher to determine.
Newer developments (not black and white)
- Causal theory
- What should be part of quantitative analysis?
- Interpretation: meaning of results depends on context
The largest part of a quantitative analysis is not about numbers.
Three possible aims to investigate associations:
- Causal inference
- Prediction
- Description
Distinguishing them makes sense, they have crucial differences in design, statistical methods,
interpretation, evaluation, role for theory/subject knowledge.
Causal inference Prediction Description
Goal Find causal effects Predict the future (or the past, or Describe patterns
- Counterfactual the current) - Identify patterns in
prediction: what if - Given what you observe the data
- Not only about what is, - No if - Matter of fact, goal
but also about what - What, given in itself
could be If you know A/B/C, what can you ® Potential starting point
say about D? for policy or further
(causal) research
Example 1. How does Netflix know what Excess mortality due to
films I like? ® watching one film coronavirus in different
makes it more likely that you will countries
like the other one too
2. Diagnosis: recognizing a
disease by the symptoms (reverse
causality a problem? ® no
because this is not about causal
inference)
1
, Combining quantitative and qualitative methods – Advanced Research Methods – Joyce Rommens
Research Why do some groups put more What kind of people will want Which groups are less open
questions value on screening than others: screening in the future? to screening?
causal inference (mediation
analysis) Background: What should the Background: In which
screening capacity in different groups could extra
Background: areas be? promotion be important?
development/testing of theory,
role of culture
Methods 1. Theory Driven 1. May be data driven: try what Bivariate associations:
2. DAG’s (exchangeability) works - Proportions/means
3. To block backdoor paths 2. Regression analysis can be /ratios per group
(randomisation, regression, used ® equation can be used for - Continuous
stratification, predictions on the individual level independent
weighting/matching) (with the first dataset, you variable ®
4. Consider blocking causal develop the regression equation correlation
paths (mediation) to predict outcomes in the coefficient
second dataset) (outdated, not
® sophisticated methods for intuitive); corm
regression may be required categories of
(prevent overfitting, the equation continuous
fits a certain dataset so well that independent
it is less suited for another variable; regression
dataset) with one
independent
No randomisation ® you do not variable
want to interfere.
No stratification ® will probably
not lead to precise predictions.
No weighting/matching ® would
require defining one exposure
Adjustment Yes! Otherwise, the results are No! This would obscure,
s affected by confounding bias remove, or increase the
due to a lack of exchangeability associations.
Results have no
interpretation anymore.
Interpretati Results have intrinsic meaning: Usually, no interest in 1. Direct, intuitive
on - Coeff represents interpretation of coeff for interpretation of
estimates of the individual predictors. Not useful: comparison of
(average) effect test whether there is an means/proportions/ratios
- Coeff in OLS association. 2. OLS coefficient
- Adjusted proportions - No clear intrinsic meaning
(average adjusted - No causal interpretation
predicted probabilities) - Coeff: ‘people with A were more
- RR, RD likely to have B/higher B, given all
- How strong is the other variables’
association?
CI Performance is crucial: evaluation
P-value may play a role of the model.
2
Lecture 6: Synthesis: quantitative and qualitative methods
Quantitative part
Objectives segment 1:
- Explain the three different goals for quantitative research (associations)
- Explain how the strategy for descriptive research differs from the strategy for causal inference
1960-2010: methodological development focused on statistical methods
- Development of new techniques
- Improvements in computers and software
- Standardized tests, ‘objectivity’
- Helpful and harmful ® we can do more and better research, but people stopped thinking
themselves and rely on the software
® Associations can be seen in the data, but it is nothing more. What do they mean is up to the
researcher to determine.
Newer developments (not black and white)
- Causal theory
- What should be part of quantitative analysis?
- Interpretation: meaning of results depends on context
The largest part of a quantitative analysis is not about numbers.
Three possible aims to investigate associations:
- Causal inference
- Prediction
- Description
Distinguishing them makes sense, they have crucial differences in design, statistical methods,
interpretation, evaluation, role for theory/subject knowledge.
Causal inference Prediction Description
Goal Find causal effects Predict the future (or the past, or Describe patterns
- Counterfactual the current) - Identify patterns in
prediction: what if - Given what you observe the data
- Not only about what is, - No if - Matter of fact, goal
but also about what - What, given in itself
could be If you know A/B/C, what can you ® Potential starting point
say about D? for policy or further
(causal) research
Example 1. How does Netflix know what Excess mortality due to
films I like? ® watching one film coronavirus in different
makes it more likely that you will countries
like the other one too
2. Diagnosis: recognizing a
disease by the symptoms (reverse
causality a problem? ® no
because this is not about causal
inference)
1
, Combining quantitative and qualitative methods – Advanced Research Methods – Joyce Rommens
Research Why do some groups put more What kind of people will want Which groups are less open
questions value on screening than others: screening in the future? to screening?
causal inference (mediation
analysis) Background: What should the Background: In which
screening capacity in different groups could extra
Background: areas be? promotion be important?
development/testing of theory,
role of culture
Methods 1. Theory Driven 1. May be data driven: try what Bivariate associations:
2. DAG’s (exchangeability) works - Proportions/means
3. To block backdoor paths 2. Regression analysis can be /ratios per group
(randomisation, regression, used ® equation can be used for - Continuous
stratification, predictions on the individual level independent
weighting/matching) (with the first dataset, you variable ®
4. Consider blocking causal develop the regression equation correlation
paths (mediation) to predict outcomes in the coefficient
second dataset) (outdated, not
® sophisticated methods for intuitive); corm
regression may be required categories of
(prevent overfitting, the equation continuous
fits a certain dataset so well that independent
it is less suited for another variable; regression
dataset) with one
independent
No randomisation ® you do not variable
want to interfere.
No stratification ® will probably
not lead to precise predictions.
No weighting/matching ® would
require defining one exposure
Adjustment Yes! Otherwise, the results are No! This would obscure,
s affected by confounding bias remove, or increase the
due to a lack of exchangeability associations.
Results have no
interpretation anymore.
Interpretati Results have intrinsic meaning: Usually, no interest in 1. Direct, intuitive
on - Coeff represents interpretation of coeff for interpretation of
estimates of the individual predictors. Not useful: comparison of
(average) effect test whether there is an means/proportions/ratios
- Coeff in OLS association. 2. OLS coefficient
- Adjusted proportions - No clear intrinsic meaning
(average adjusted - No causal interpretation
predicted probabilities) - Coeff: ‘people with A were more
- RR, RD likely to have B/higher B, given all
- How strong is the other variables’
association?
CI Performance is crucial: evaluation
P-value may play a role of the model.
2
