Andreas Flache Manu Muñoz-Herrera Explanation and Prediction Lecture Week 2 - Application of Theories Block A 2012/2013
May 12, 2015
Andreas Flache
Manu Muñoz-Herrera
Explanation and PredictionLecture Week 2 - Application of Theories
Block A 2012/2013
Brief Summary:Where have all the criminals gone?Chapter from the book: Freakonomics
Steven Levitt Stephen Dubner
Why was there a drop in violent crime in the U.S. since the 1990’s?
There was a drop in violent crimes in the U.S. since 1990
Observe a phenomenon Speculate about it
There were multiple explanations given, ranging from police enforcement to increase in economic welfare
Not all speculated causes explained the drop. Those “valid” explanations accounted for 50%
Ask if the implications correct Deduce other results
If an increase in welfare decreases violent crime, then we should see this in other states or time points
Surprising Explanation: Abortion
Legalization of abortion leads to less unwanted children
Unwanted children are more likely to become criminals
The legalization of abortion leads to less criminals (a drop in crime).
HOW can we test this?
Sociological research questionsExplanation & Prediction
We want to explain/predict social phenomena
You formulate why-questionswhen you seek to explain a social phenomenon.
These questions are related to predictions about social phenomena.
Why was there a revolution in Eastern Germany in 1989?
Why was there in the late 19th century more suicides among protestants than among catholics?
Will there be a revolution in Iran?
Will there be more suicides among protestants than among catholics in the Netherlands in 2014?
Aims of the lecture
In this lecture we will learn:
How to explain and/or predict phenomena.
Which criteria define an adequate explanation or prediction?
What is the problem of induction?
Frequently encountered problems with explanations
Checklist.
Part 1: How to explain/predict social phenomena?
Starting ExampleRevolution in Eastern Germany
Example:Why was there a revolution in Eastern Germany in 1989?
Summer 1989: The East German government praised (on t.v.) the decision to use violence against the Tiananmen Square protesters.
Example:Why was there a revolution in Eastern Germany in 1989?
Sept. 4, 1989: Demonstrations began at St. Nicholas church in Leipzig.
Example:Why was there a revolution in Eastern Germany in 1989?
Oct. 2, 1989: Eric Honecker issued a shoot to kill order; huge police militia, Stasi, and work-combat troop presence.
Example:Why was there a revolution in Eastern Germany in 1989?
Oct. 9, 1989: Biggest peaceful demonstration. Military surrounded the demonstration but did not take action.
Example:Why was there a revolution in Eastern Germany in 1989?
Oct. 18, 1989: Honecker had to resign.
Example:Why was there a revolution in Eastern Germany in 1989?
Nov. 9, 1989: Fall of the Berlin wall.
Example:Why was there a revolution in Eastern Germany in 1989?
Oct. 3, 1990: German reunification.
To explain the revolution, many (sub)questions need to be answered. For example:
Why did people do this?
Why did officials eventually not order to shoot to the protestors?
Why did Honecker decide to resign?
Why didn’t the soviet army intervene? (it had happened before)
However, one of the most interesting questions is:
Why did so many people participate in the protest in Leipzig?
How can we explain this phenomenon?
Note that our explanandum is a so called singular phenomenon: We seek to explain one protest not all protests in the world.
Participation in Monday Demonstrations
Source: Table 1 from Braun, Norman. 1995. Individual Thresholds and Social Diffusion. Rationality and Society 7:167-182.
0
125,000
250,000
375,000
500,000
Sept25 Oct2 Oct9 Oct16 Oct23 Oct30 Nov6
Participants
Let’s give it a first try
Question. Why did many people participate in the protests?
Answer. Many people were very dissatisfied with the living conditions in Eastern Germany and therefore decided to protest.Problem. How do we know that it was the dissatisfaction with the living conditions? People were dissatisfied with many things (no freedom of press, repression).Defense. But we observed that in 1989 many people in Leipzig were dissatisfied with the living conditions.Still. Sure, but we observed many things this year (only in 1989 Steffi Graf and Boris Becker won Wimbledon). Why is this not the reason?
What are we doing here?
We try to formulate the “laws” or “processes” that could have produced the phenomenon. We express these in theories... and test the theories.
Hempel & Oppenheim, pg. 153: Theories consist of general statements (laws, or “law-like”) which are not restricted to certain objects, or certain dates or places.
The Covering Law ModelGeneral method of explaining and predicting
Alternative names: Deductive-nomological model (D-N Model) Subsumption theory Hempel’s model Hempel-Oppenheim model Popper-Hempel model of explanation
Carl Gustav Hempel1905-1997
The covering law model
Question. Why did many people participate in the protests?
Answer.In general, if many people in a society are dissatisfied with the living conditions, then they will protest.
In Leipzig 1989, many people were dissatisfied with the living conditions.
In Leipzig 1989, many people protested.
In general, if many people in a society are dissatisfied with the living conditions, then they will protest.
In Leipzig 1989, many people were dissatisfied with the living conditions.
In Leipzig 1989, many people protested. Explanandum
Explanans
The line indicates that explanandum follows from the explanans
Explanandum (E). Statement E that we seek to explain. Singular statement (“existential”: exists in concrete time-place context).
Explananda (L1... + C1...). Statements that explain E. Consists of at least one general statement, or law, (L1) and at least one
singular statement (C1), also called “condition”. The latter is also a singular statement.
The general structure of the covering law model
In general, if many people in a society are dissatisfied with the living conditions, then they will protest.
In Leipzig 1989, many people were dissatisfied with the living conditions.
In Leipzig 1989, many people protested.
Law: It is true for all x: if Dx, then PxSingular statement: Da
Singular statement: Pa
Law: It is true for all x: if Dx, then PxSingular statement: Da
Singular statement: Pa
a is a subset of x (individuals who lived in 1989 in Leipzig are “people”). Either the then-component of the law is identical to the
explanandum or it is a subset of the explanandum. For instance, this is not correct:
In general, if many people in a society are dissatisfied with the living conditions, then they will protest.
In Leipzig 1989, many people were dissatisfied with the living conditions.
In Leipzig 1989, many people signed a petition.
Signing a petition is a form of protest but the inference is not correct, because people might have protested in another way.
Second example:
In general, if many people in a society are dissatisfied with the living conditions, then they will protest.
In Leipzig 1989, many people were dissatisfied with the living conditions.
In Leipzig 1989, many people were politically active.
Protest is one form of political action (among e.g., voting). The law includes that there will be more protest. As this is a political action, also political action will increase.
Thus, the explanandum follows from the explanans. Political action
Protest
Signing a petition
A classic sociological exampleDurkheim’s theory of suicide
Emile Durkheim1858-1917
Example:Durkheim’s theory of suicide
The explanandum:(Between 1867 and 1875 in Bavaria), the larger the proportion of catholics in a province, the less likely people in that province were to commit suicide.
0
50
100
150
200
<50% 50%-90% >90%
Suicide rate per milliollion Inhabitants
Source: Finlay, W. “What is sociology”, http://www.arches.uga.edu/%7Ewfinlay/SOCI1101H.htm, Retrieved November 2005.
Example:Durkheim’s theory of suicideExplanandum (E). This explanandum has a bit different structure. “The more Kx, the less
Sy”. This is what you often encounter in real explanation problems. In principle, the same methods can be used for this. Here is a
reconstruction of Durkheim’s explanation:
The more the members of a religious group are integrated in this group, the less likely they commit suicide (L1)
In Bavaria, between 1867 and 1875: Catholics were more integrated in their religious groups than protestants (C1).
In Bavaria, between 1867 and 1875: catholics were less likely to commit suicide than protestants.
Explanandum
Explanans
Part 2: Conditions of adequacy
These conditions need to be met. Otherwise, the explanation is not “sound”.
They have been formulated by Hempel and Oppenheim.
Many further conditions have been added. The following are minimal conditions.
Condition 1:
The explanandum must be a logical consequence of the explanans
This means: If all statements in the explanans are true, then the explanandum
must be true too.
Next week, you will learn how to check this condition.
Condition 2:
The explanans must contain at least one general law and at least one singular statement&Both must be actually required to logically derive the explanandum
Note that the singular statement is not needed when we want to explain a law. For example, derive a more specific law from a more general law.
Example: Integration -> people follow all norms more Integration -> people follow the norm “do not commit suicide” more
Condition 3:
The explanans must have empirical content
Example: The law “if there is anomie in a society, then people will be truly unhappy” is
not testable, because “anomie” and “truly unhappy” are (yet) undefined.
This means: The explanans must inform about reality. In other words, it must
consist of or generate testable statements (and could be wrong).
Condition 4:
All statements of the explanans must be true
This condition is debated a lot: A “law” is universal and therefore it is hard to prove it is true.
Milton Friedman1912-2006
Nobel 1976: Monetarism
Explanans can be wrong,what counts is a correct
prediction
But don’t we strive for good explanations and not just for good predictions?
Simplicity + Fertility + Surprise + Empirical Support
Probabilistic Laws
Probabilistic Laws
Until now: We have focused on deterministic laws. In practice,
however, most theories make probabilistic statements.
Example:
If in a society many people are dissatisfied with their living conditions, a revolution is likely
In 1989 in the GDR many people were dissatisfied with their living conditions.
In 1989 in the GDR, a revolution was likely
What if there had not been a revolution in 1989 in the GDR? Would we consider the law to be wrong? No, because the law is about all possible societies, but our study
focused on just one society. Thus, one case does not prove that it is not more likely to have a revolution if many people are dissatisfied. To avoid this problem, scientists (implicitly) use reformulations of
the law like this: If we observe a large number of societies, we will see more revolutions in societies in which more people are dissatisfied.
If in a society many people are dissatisfied with their living conditions, a revolution is likelyIn 1989 in the GDR many people were dissatisfied with their living conditions.
In 1989 in the GDR, a revolution was likely
Problem 1 with probabilistic laws
Based on a weighted scale: Share of the population that is
under 25 (35%) Number of years the government
has been in power (15%) Corruption (15%) Lack of democracy (15%) GDP per person (10%) Index of censorship (5%) Absolute number of people
younger than 25 (5%)
Example: The Arab Spring(not just for explanation, but also for prediction)
Source: The Economist (Feb. 12th, 2011)Aim: “to predict where the scent of jasmine may spread next” (after Tunisia, before Egypt)
Is this a valid argument? No, because the law does not allow for exceptions and this case
might be one. Thus, probabilistic laws are problematic when we use them to
explain very specific events.
In cities where many government officials lie, people are less likely to protest
There were more government officials in Berlin than in Leipzig
In Berlin, fewer people protested
Problem 2 with probabilistic lawsAssume that we want to explain something very specific. For instance, why did more people protest in Leipzig than in Berlin?
A single counterexample falsifies a general statement. Especially, in the social sciences there is no law without an exception. Thus, do we have laws that we can use? This is an important problem, and there is no solution to it. But this is an important reason why in the practice of research we often work with probabilistic laws, or seek for more general theories that explain the exception.
The barometer is falling rapidly
Whenever the barometer is falling rapidly, a storm is approaching
A storm is approaching
Problems with the covering law model
Asymmetry Problem
The explanandum follows logically correct from the explanans. Nevertheless, a falling barometer does not cause a storm.
Part 3: InductionThus, is there a better method than deduction?
The inductive approachSome social scientists argue that an inductive approach is better.
During the French Revolution, many people were very dissatisfied with their living conditions.
In Leipzing 1989, many people were very dissatisfied with their living conditions
It is always true that there will be protest if people were very dissatisfied with their living conditions.
...
John Stuart Mill 1806-1873
Inductive reasoning means that one generalizes from several observations to general laws
David Humes1711-1776
David Humes, however, discovered a problem with inductive reasoning
The covering law model is based on the deduction principle. It tells us that if all statements of the explanans are true, then the explanandum must be true as well. (This follows from logic- see next week).
The deduction principle is a general statement (a law) which is always true.
There is, however, no induction principle. to arrive at one, one would need to use induction (= infer the general principle from several single cases). Therefore, however, you need an induction principle. To get this, in turn, you need an induction principle.
Induction implies an infinite regress. This makes it problematic.
Predictions
Based on a weighted scale: Share of the population that is
under 25 (35%) Number of years the government
has been in power (15%) Corruption (15%) Lack of democracy (15%) GDP per person (10%) Index of censorship (5%) Absolute number of people
younger than 25 (5%)
Example: Predicting Revolutions/ The Shoe-Thrower’s index
Source: The Economist (Feb. 12th, 2011)Aim: “to predict where the scent of jasmine may spread next” (after Tunisia, before Egypt)
Predictions: This is the right way to do it.Predictions are very similar to explanations, but they are not the same.
In general, if many people are dissatisfied with the living conditions, then they will protest.
In Leipzig 1989, many people were dissatisfied with the living conditions.
In Leipzig 1989, many people protested.
In general, if many people are dissatisfied with the living conditions, then they will protest.
Due to an embargo in Iran, many people’s satisfaction with living costs will be low
There will be a protest in Iran
Explanation
Prediction
We want to find this
We know this
We know this
We want to find this
Most laws do not specify how long one has to wait until the prediction will become true and for how long one can observe it.
Thus, the prediction might be correct, but when we test it at the wrong moment, we might not notice it.
Predictions often implicitly assume that all other conditions will not change (ceteris paribus).
This may be wrong. For instance, the government of Iran might increase repression.
Often, the singular statements (conditions) need to be predicted as well (e.g., there will be an embargo).
This can lead to an infinite regress. Self-fulfilling (e.g., bank run) and self-killing (e.g., Obama will win
the elections) prophecies.
Problems with predictions
Often predictions are data driven: Scientists measure a development in the past (e.g., size of world population) and extrapolate developments in the future. This approach presumes that observed trends will continue for ever.
This can be very wrong. Remember, for instance, the development of crime rates in the U.S.
Predictions without laws?
Part 4: Check List
Check List
Try to include in the explanans theoretically plausible or empirically supported statements (=assumptions). This is key if the prediction depends very much on this particular assumption.
Of course, statements which have not been tested and which are not plausible can be true. So allow room for speculation Typical assumptions which appear innocent but often are crucial:
linearity, normal distribution, huge groups. Use one word to describe a concept. Avoid synonyms, or at least define
them explicitly. Never describe more than one phenomenon with one concept.
E.g., “socially integrated in a group”, “attached to the group”, “strongly identifying with a group”, could all mean the same in theory (that can be confussing).
when you develop an explanation or a prediction, take care of the following things (in addition to checking the adequacy conditions.
Check List (II) Formulate the laws as general as possible (see also next week)
E.g., All students need to eat All humans need to eat All mammals need to eat
However, sometimes generalization does not make sense
All students follow the lecture All mammals follow the lecture
If a statement is true, it does not imply that the generalized statement is true as well (that would be induction).
Check List (III)
However, the more general a statement is the more empirical content it typically has. That is: the more testable statements you can derive from it.
For instance, with a general theory of political protest, we can not only test the theory of Leipzig 1989, but also France 1789 and for Iran 2013, etc.
If you do not find the right laws, try to derive them from other, more general laws.
If the explanandum is complex, break it down to subproblems and tackle them one-by-one (e.g., revolution -> protest).
How can we test Durkheim’s theory of suicide and social integration?
L1: The more people are integrated in social groups, the less likely they commit suicide.
We need to find a condition (singular statement) about a difference in integration in social group between people
Here is a suggestion:
C1: In the Netherlands, in the late 20th century, married people are more integrated in social groups than are unmarried people.
L1C1- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - E1: In the Netherlands, in the late 20th century, married people will less likely commit suicide than unmarried people
A final example (a bit more complicated)
The testE1: In the Netherlands, in the late 20th century, married people will less likely commit suicide than unmarried people
Bron. Ultee, Aarts, Flap. 1996. Sociologie: vragen, uitspraken, bevindingen. Groningen: Wolters Nordhoff.
Why ceteris paribus is so importantE1: In the Netherlands, in the late 20th century, married people will less likely commit suicide than unmarried people
To be precise, all we can really plausibly assume is that more integrated people are less likely to commit suicide than are people who otherwise have the same characteristics, or
L1: The more people are integrated in social groups, the less likely they commit suicide
Ceteris paribus = all other things being equal
But are all other things equal between married and unmarried people with regard to characteristics that matter for suicide?
Another testE1: In the Netherlands, in the late 20th century, married people will less likely commit suicide than unmarried people, all other things being equal
Bron. Ultee, Aarts, Flap. 1996. Sociologie: vragen, uitspraken, bevindingen. Groningen: Wolters Nordhoff.
The testE1: In the Netherlands, in the late 20th century, married people will less likely commit suicide than unmarried people
Bron. Ultee, Aarts, Flap. 1996. Sociologie: vragen, uitspraken, bevindingen. Groningen: Wolters Nordhoff.
On Nestor, you find the assignment and the reading material for Thursday.