Overview
Lecture 1: logical positivists
Logically verifiable
Empirically verifiable
Scientific if: logically consistent & all empirical terms operationalized
Operationalization
Demarcation
Assessment: context of discovery, context of justification
DN-model
Symmetry thesis
Law-like proposition
Verification of laws: humean problem of induction -> instrumentalism, confirmationism
Lecture 2:
Econometrics, positivism in econometrics
Keynes-Tinbergen
Keynes’ criticism of Tinbergen’s econometric model:
1. Tinbergen does not test but measures
2. Tinbergen’s measuring only possible is all causalities known (complete model)
Haavelmo
Probability approach
Economic laws: found if can find model complete enough to stabilize parameters
Cannot test, only measure. Goals is measuring by means of probabilities (not exact values)
No way to tell whether to keep variable in or leave out
What to do? Wait to become important & then include in model? Always repairing model after the fact (not good) or start with as comprehensive model as
possible, how to keep model manageable see next
NBER and koopmans: measurement-without-theory debate
How to keep model manageable
Koopmans criticized NBER for thinking stats is all there is, says NBER looking at stats in hope to find irregularities for model specification. WRONG. First start
with model specification and then do testing (base on what can be expected -> model specification -> go to data).
NBER says theory based on observation alone (does not work), theory tells us where to look then can test hypotheses. Theory input & output of observation
-> introspection (= think of behavior by putting themselves in their shoes)
Problems of econometrics: (i) no laboratory, so no certainties re cause and effect (ii) more variables thus more certainty, invariance, and complexity (iii) start
big or start on small basis and expand model -> complexity unavoidable. How to deal with this complexity see next.
Friedman: methodology if positive economics
Startingpoint: models because too complex -> test eco models against naïve models -> naïve models outperform
Validity of models: look for models that outperform naïve models, test of model is how well it predicts. Keep simple and general.
How to simplify? Use simple models (not lot variables) that predict well.
A model is scientific if it predicts well on well-specified domains (domain=simplified element of reality). Limit set of variables and make good predictions on
that specific domain.
Anti-realisticness: realism of assumption doesn’t matter, only need descriptions that predicts well. (tree-leaf). Causal relationship is irrelevant.
According to Musgrave, this is not what Friedman tried to say:
Model assumptions: (i)domain: expected factor absent, so is used to specify domain (ii)negligibility: factor that could be expected to affect phenomenon
actually no effect (iii)heuristic: if factor considered negligible, in order to simplify ‘logical’ development of theory
- what Friedman is doing: come up with specific domain and then say on this domain certain variables are most relevant (if all, too much complexity).
- Negligibility assumption limit yourself to those things that are important and neglect the rest on that specific domain, Its not a call for anti-realisticness but
call for preventing too much realism -> prevent too much complexity. This is what Musgrave is saying.
- friedman: a model is scientific if it predicts well on well-specified domains. Predicts well if:
1. limit ourselves to most relevant variables on that domain or
2. if can argue that certain simulacrum will give same results but is easier to model (gives right predictions)
D-N model only used for predictions, so testing law directly impossible, so just use above options.
Laws themselves not scientific, intrumentalistic: if laws predict right, scientific. Schlick says never scientific because can never be certain about laws. Critique
friedman -> Samuelson
Samuelson: critique of the F-twist
Hope he misunderstood friedman, that’s why called F-twist (anti-realisticness, as long as predicts right)
Truth table for implication: is how samuelson arguid against friedman. He said, this is what friedman is on about.
If have false premise and draw true conclusion from it, its logically consistent(scientific). This is what Samuelson understands from what friedman is doing.
Accepting false premies that have true consequences are scientific.
To dismiss this idea, comes up with table of equivalence (because still should inform theories as well.
Truth table for equivalence: premise leads to consequence, but consequence also to premise (need to correlate)
If have false premise (observation), draw truthful conclusion (consequence), make mistake in logic (logically inconsistent, not scientific). My predict right but
empirically wrong.
descriptivism (=set of assumptions, theory, consequences, and they should all move into one another). Consequence of this: as samuelson put it, scientists
never explain behavior -> there are NO laws -> no DN explanations or predictions, only descriptions -> never scientific.
Samuelson comes full circle back to logical positivism saying: forget it, then we don’t explain or predict I’d rather have that than accept nonsense of friedman
Lecture 3: Popper’s falsificationism
Logic der Forshung his answer to humean induction problem
LP: laws problematic, popper’s solution: falsification
Falsification: try to refute laws
Something is true until proven otherwise (burden of proof)
Burden of proof has implications:
1. laws are back (basis of science): start with law, only then can try and refute what we found
2. scientific knowledge is fallible: talk like “as far as we know there might be..”, not: “science proves that is not”
Science cannot prove, only use hypotheses.
3. scientists should aim for refutations rather than verification & welcome criticism
4. theories have to be risky: at risk of being disproven. Start out with theory that is based on no observation.
Non-risky theory: acc to popper pseudo-science immune theories, as soon talk about subconscious/suppressed feelings, its hard to disprove.
, Key difference what popper and LP consider scientific: because to LP, claim like dragons exist as long as operationalized dragons and we’ve actually had one
observation, is a scientific statement.
To popper, statement that something exists can be disproven, there isn’t one observation that can disprove existence of something.
Acc tp popper, science should be a general statement. So scientific statement would be to popper: god does not exist. Then you have to operationalize god,
and someone can prove me wrong
This and that exists: Scientific to logical positivists, non-scientific to popper. This or that does not exist: non-scientific to logical positivists, scientific to popper
Testing theories popper style: (i) logical consistency (min. requirement) (ii) falsifiabilitt (demarcation criterion) (iii) empirical consequences (predictions) (iv)
empirical content (falsifiers)
The more specific and general a theory becomes, the larger its empirical content. And the better the theory is acc. to popper. The higher the empirical
content the higher the potential falsifiers. So empirical content rises as theory more specific about greater nr. of phenomena. Besides this, comparisons may
be based on degree of corroboration. Popper uses corroboration, LP uses verification, but it same thing (confirmation)
there is development of knowledge but its not a straight line. Whereas LP were in sense saying knowledge is a building and we start w/certain building
blocks. Build upon logically, building of knowledge is increasing.
Popper was saying: you put up whole building and then invite people to break it down. if successful in breaking down, you set up new stronger building.
Stronger building is building with more empirical content. More precise predictions about more phenomena.
Theory ladenness (=observation done in light of a theory). w/popper, observations are theory lade, because theory precedes observation.
Immunzing stratagems a posteriori: ways to save ass after criticism (not scientific tactics, but human)
1) re-classify: “if its black cant be a swan, I am the expert.” You’re deflecting criticism (not being scientific, bad)
2) adjust underlying theories: “black if form of darkwhite, im not wrong, you are”
3) change domain: “Australia is crazy, this not part of domain, only EU because that’s normal country” so my theory doesn’t apply to Australia, they are
crazy.
Immunizing stratagems a priori: ceteris paribus (all other things being equal)
- if you say CP: So if you beforehand say all other things equal, then you can always point to something that isn’t in your theory and has changed and
therefore your theory is still right, so then you’re immunizing even before you criticize. You’re already deflecting criticism before that criticism is even
forthcoming (not scientific)
- if you specify: you can still be proven correct or incorrect. (scientific)
Duhem-Quine
They said to popper: with your immunizing strategems, and forbidding immunizing strategemens wholesale, you’re neglecting whole part of development in
the sciences. because when people engage in immunization, they are actually improving the theory
Measurement errors, by miscalculation, check measurements again to improve theory
Example walvis, which is fish, but can breath above water (theory was couldn’t breath on land). If popperation you say fish cant drown, so observation of
whale proves theory of fishes is wrong. By immunization: biologists said well, perhaps wale isn’t actually a fish, and continued on that track. Immunization is
theory improvement, because this way we understand wales much better and understand fish better, but we’re limiting the domain and we’re reclassifying,
saying: wale is not fish, in this case its an improvement and so improves theories. Because if we do not know what is at fault, immunizing strategies may
actually improve the theory.
Main diff: when theory appears falsified then popper says: the whole theory is debunked and we should start a new and replace it with a theory with larger
empirical content. Whereas friedman says: if theory debunked then we’ve applied it to wrong domain, so for friemdna theory always comes from
immunization (changing the domain in response to falsifying instances) Popper was against this. Furthermore they both say theory got to be improbable in
that sense that its risky and predictions may well or may not hold
Social laws: you can’t do experiments(difficult) can never ascertain one factor causes another to change. May be related to factors we haven’t taken into
account. Many factors, making it very hard to find. What he adds to this problem is reflexivity (=if laws are found, our behavior changes, voiding the law
social and historical inescapability) Example: if we have theory about stock market, people change behavior immediately to take advantage of this new
theory. Voiding the theory. Known as reflexivity.
Lecture 4: Kuhn and Lakatos
Kuhn
How is it possible that something that seems absurd now, was taken serious for more than 1700 years? This is puzzle Kuhn tried to solve. For answers he
turned to sociology, saying that normal science takes place in scientific community with specific norms and values. It is because you normalized into a certain
scientific community that you see things in a certain light. Cannot see world in different light anymore.
“you’ll see the light in time”
Normal science in scientific community characterized by norms and values regarding
1. research domain: what is economics and what is not
2. research methods: if have domain, and say what what economist study, how do we study this? Relates to modeling, math, optimizing
3. criteria for good science: if predict well & good statistical firt w/data.
Disciplinary matrix: in order to figure out what shared norms/values are of scientific community, we should look for characteristics of their disciplinary matrix
1. symbolic generalization: termsthat are common toparticular scientific field,meaningless to other scientific field
2. metaphysics: whats your go-to way to structure your thoughts? Supply and demand (understanding this)
3. paradigm: start of new world view, thinking back to analyze markets.
4. values: when is model/theory good? Good/bad science.
Educating and policing:
- education by standard examples: first learn to see world from our perspective, when you can, you can criticize us. You become member of scientific
community because no longer able to think critically. You’ve been convinced by examples that were used.
- metaphors: view businesses as machine that can be finetuned, functions optimally.
- these are kind ways, not kind ways professor saying if you don’t believe me you will fail exam
- policing through invisible college: you got to make sure you don’t make anyone up in hierarchy angry if don’t publish you don’t get promoted etc. hierarchy
makes us all fall in line and not criticizing.
- All this diff from what popper said: good scientists needs to be open to criticism. You are accepted as GOOD scientists if you’re not open to criticism, you’re
climbing the ladder then.
Gestalt: what you see in it. All interpretations go but we only accept 1. (world-view)
Lecture 1: logical positivists
Logically verifiable
Empirically verifiable
Scientific if: logically consistent & all empirical terms operationalized
Operationalization
Demarcation
Assessment: context of discovery, context of justification
DN-model
Symmetry thesis
Law-like proposition
Verification of laws: humean problem of induction -> instrumentalism, confirmationism
Lecture 2:
Econometrics, positivism in econometrics
Keynes-Tinbergen
Keynes’ criticism of Tinbergen’s econometric model:
1. Tinbergen does not test but measures
2. Tinbergen’s measuring only possible is all causalities known (complete model)
Haavelmo
Probability approach
Economic laws: found if can find model complete enough to stabilize parameters
Cannot test, only measure. Goals is measuring by means of probabilities (not exact values)
No way to tell whether to keep variable in or leave out
What to do? Wait to become important & then include in model? Always repairing model after the fact (not good) or start with as comprehensive model as
possible, how to keep model manageable see next
NBER and koopmans: measurement-without-theory debate
How to keep model manageable
Koopmans criticized NBER for thinking stats is all there is, says NBER looking at stats in hope to find irregularities for model specification. WRONG. First start
with model specification and then do testing (base on what can be expected -> model specification -> go to data).
NBER says theory based on observation alone (does not work), theory tells us where to look then can test hypotheses. Theory input & output of observation
-> introspection (= think of behavior by putting themselves in their shoes)
Problems of econometrics: (i) no laboratory, so no certainties re cause and effect (ii) more variables thus more certainty, invariance, and complexity (iii) start
big or start on small basis and expand model -> complexity unavoidable. How to deal with this complexity see next.
Friedman: methodology if positive economics
Startingpoint: models because too complex -> test eco models against naïve models -> naïve models outperform
Validity of models: look for models that outperform naïve models, test of model is how well it predicts. Keep simple and general.
How to simplify? Use simple models (not lot variables) that predict well.
A model is scientific if it predicts well on well-specified domains (domain=simplified element of reality). Limit set of variables and make good predictions on
that specific domain.
Anti-realisticness: realism of assumption doesn’t matter, only need descriptions that predicts well. (tree-leaf). Causal relationship is irrelevant.
According to Musgrave, this is not what Friedman tried to say:
Model assumptions: (i)domain: expected factor absent, so is used to specify domain (ii)negligibility: factor that could be expected to affect phenomenon
actually no effect (iii)heuristic: if factor considered negligible, in order to simplify ‘logical’ development of theory
- what Friedman is doing: come up with specific domain and then say on this domain certain variables are most relevant (if all, too much complexity).
- Negligibility assumption limit yourself to those things that are important and neglect the rest on that specific domain, Its not a call for anti-realisticness but
call for preventing too much realism -> prevent too much complexity. This is what Musgrave is saying.
- friedman: a model is scientific if it predicts well on well-specified domains. Predicts well if:
1. limit ourselves to most relevant variables on that domain or
2. if can argue that certain simulacrum will give same results but is easier to model (gives right predictions)
D-N model only used for predictions, so testing law directly impossible, so just use above options.
Laws themselves not scientific, intrumentalistic: if laws predict right, scientific. Schlick says never scientific because can never be certain about laws. Critique
friedman -> Samuelson
Samuelson: critique of the F-twist
Hope he misunderstood friedman, that’s why called F-twist (anti-realisticness, as long as predicts right)
Truth table for implication: is how samuelson arguid against friedman. He said, this is what friedman is on about.
If have false premise and draw true conclusion from it, its logically consistent(scientific). This is what Samuelson understands from what friedman is doing.
Accepting false premies that have true consequences are scientific.
To dismiss this idea, comes up with table of equivalence (because still should inform theories as well.
Truth table for equivalence: premise leads to consequence, but consequence also to premise (need to correlate)
If have false premise (observation), draw truthful conclusion (consequence), make mistake in logic (logically inconsistent, not scientific). My predict right but
empirically wrong.
descriptivism (=set of assumptions, theory, consequences, and they should all move into one another). Consequence of this: as samuelson put it, scientists
never explain behavior -> there are NO laws -> no DN explanations or predictions, only descriptions -> never scientific.
Samuelson comes full circle back to logical positivism saying: forget it, then we don’t explain or predict I’d rather have that than accept nonsense of friedman
Lecture 3: Popper’s falsificationism
Logic der Forshung his answer to humean induction problem
LP: laws problematic, popper’s solution: falsification
Falsification: try to refute laws
Something is true until proven otherwise (burden of proof)
Burden of proof has implications:
1. laws are back (basis of science): start with law, only then can try and refute what we found
2. scientific knowledge is fallible: talk like “as far as we know there might be..”, not: “science proves that is not”
Science cannot prove, only use hypotheses.
3. scientists should aim for refutations rather than verification & welcome criticism
4. theories have to be risky: at risk of being disproven. Start out with theory that is based on no observation.
Non-risky theory: acc to popper pseudo-science immune theories, as soon talk about subconscious/suppressed feelings, its hard to disprove.
, Key difference what popper and LP consider scientific: because to LP, claim like dragons exist as long as operationalized dragons and we’ve actually had one
observation, is a scientific statement.
To popper, statement that something exists can be disproven, there isn’t one observation that can disprove existence of something.
Acc tp popper, science should be a general statement. So scientific statement would be to popper: god does not exist. Then you have to operationalize god,
and someone can prove me wrong
This and that exists: Scientific to logical positivists, non-scientific to popper. This or that does not exist: non-scientific to logical positivists, scientific to popper
Testing theories popper style: (i) logical consistency (min. requirement) (ii) falsifiabilitt (demarcation criterion) (iii) empirical consequences (predictions) (iv)
empirical content (falsifiers)
The more specific and general a theory becomes, the larger its empirical content. And the better the theory is acc. to popper. The higher the empirical
content the higher the potential falsifiers. So empirical content rises as theory more specific about greater nr. of phenomena. Besides this, comparisons may
be based on degree of corroboration. Popper uses corroboration, LP uses verification, but it same thing (confirmation)
there is development of knowledge but its not a straight line. Whereas LP were in sense saying knowledge is a building and we start w/certain building
blocks. Build upon logically, building of knowledge is increasing.
Popper was saying: you put up whole building and then invite people to break it down. if successful in breaking down, you set up new stronger building.
Stronger building is building with more empirical content. More precise predictions about more phenomena.
Theory ladenness (=observation done in light of a theory). w/popper, observations are theory lade, because theory precedes observation.
Immunzing stratagems a posteriori: ways to save ass after criticism (not scientific tactics, but human)
1) re-classify: “if its black cant be a swan, I am the expert.” You’re deflecting criticism (not being scientific, bad)
2) adjust underlying theories: “black if form of darkwhite, im not wrong, you are”
3) change domain: “Australia is crazy, this not part of domain, only EU because that’s normal country” so my theory doesn’t apply to Australia, they are
crazy.
Immunizing stratagems a priori: ceteris paribus (all other things being equal)
- if you say CP: So if you beforehand say all other things equal, then you can always point to something that isn’t in your theory and has changed and
therefore your theory is still right, so then you’re immunizing even before you criticize. You’re already deflecting criticism before that criticism is even
forthcoming (not scientific)
- if you specify: you can still be proven correct or incorrect. (scientific)
Duhem-Quine
They said to popper: with your immunizing strategems, and forbidding immunizing strategemens wholesale, you’re neglecting whole part of development in
the sciences. because when people engage in immunization, they are actually improving the theory
Measurement errors, by miscalculation, check measurements again to improve theory
Example walvis, which is fish, but can breath above water (theory was couldn’t breath on land). If popperation you say fish cant drown, so observation of
whale proves theory of fishes is wrong. By immunization: biologists said well, perhaps wale isn’t actually a fish, and continued on that track. Immunization is
theory improvement, because this way we understand wales much better and understand fish better, but we’re limiting the domain and we’re reclassifying,
saying: wale is not fish, in this case its an improvement and so improves theories. Because if we do not know what is at fault, immunizing strategies may
actually improve the theory.
Main diff: when theory appears falsified then popper says: the whole theory is debunked and we should start a new and replace it with a theory with larger
empirical content. Whereas friedman says: if theory debunked then we’ve applied it to wrong domain, so for friemdna theory always comes from
immunization (changing the domain in response to falsifying instances) Popper was against this. Furthermore they both say theory got to be improbable in
that sense that its risky and predictions may well or may not hold
Social laws: you can’t do experiments(difficult) can never ascertain one factor causes another to change. May be related to factors we haven’t taken into
account. Many factors, making it very hard to find. What he adds to this problem is reflexivity (=if laws are found, our behavior changes, voiding the law
social and historical inescapability) Example: if we have theory about stock market, people change behavior immediately to take advantage of this new
theory. Voiding the theory. Known as reflexivity.
Lecture 4: Kuhn and Lakatos
Kuhn
How is it possible that something that seems absurd now, was taken serious for more than 1700 years? This is puzzle Kuhn tried to solve. For answers he
turned to sociology, saying that normal science takes place in scientific community with specific norms and values. It is because you normalized into a certain
scientific community that you see things in a certain light. Cannot see world in different light anymore.
“you’ll see the light in time”
Normal science in scientific community characterized by norms and values regarding
1. research domain: what is economics and what is not
2. research methods: if have domain, and say what what economist study, how do we study this? Relates to modeling, math, optimizing
3. criteria for good science: if predict well & good statistical firt w/data.
Disciplinary matrix: in order to figure out what shared norms/values are of scientific community, we should look for characteristics of their disciplinary matrix
1. symbolic generalization: termsthat are common toparticular scientific field,meaningless to other scientific field
2. metaphysics: whats your go-to way to structure your thoughts? Supply and demand (understanding this)
3. paradigm: start of new world view, thinking back to analyze markets.
4. values: when is model/theory good? Good/bad science.
Educating and policing:
- education by standard examples: first learn to see world from our perspective, when you can, you can criticize us. You become member of scientific
community because no longer able to think critically. You’ve been convinced by examples that were used.
- metaphors: view businesses as machine that can be finetuned, functions optimally.
- these are kind ways, not kind ways professor saying if you don’t believe me you will fail exam
- policing through invisible college: you got to make sure you don’t make anyone up in hierarchy angry if don’t publish you don’t get promoted etc. hierarchy
makes us all fall in line and not criticizing.
- All this diff from what popper said: good scientists needs to be open to criticism. You are accepted as GOOD scientists if you’re not open to criticism, you’re
climbing the ladder then.
Gestalt: what you see in it. All interpretations go but we only accept 1. (world-view)
