# Hypothetico-deductive method

Scientific method whereby science should set up testable hypotheses and then try to falsify them, rather than trying to confirm them directly by accumulating favourable evidence.

Introduced by the English scholar William Whewell (1794-1866) and developed especially by the Austrian philosopher Karl Raimund Popper (1902-1994).

Those hypotheses which – despite severe tests – survive unfalsified are thereby confirmed for Whewell; Popper goes further and says they are merely ‘corroborated’, a notion which is supposed to avoid the logical invalidity associated with induction.

Also see: falsificationism, inductivism, deductivism, improbabilism

Source:
K R Popper, The Logic of Scientific Discovery (1959); unrevised German original 1934

## Example

One example of an algorithmic statement of the hypothetico-deductive method is as follows:[1]

1. Use your experience: Consider the problem and try to make sense of it. Gather data and look for previous explanations. If this is a new problem to you, then move to step 2.
2. Form a conjecture (hypothesis): When nothing else is yet known, try to state an explanation, to someone else, or to your notebook.
3. Deduce predictions from the hypothesis: if you assume 2 is true, what consequences follow?
4. Test (or experiment): Look for evidence (observations) that conflict with these predictions in order to disprove 2. It is a logical error to seek 3 directly as proof of 2. This formal fallacy is called affirming the consequent.[2]

One possible sequence in this model would be 1234. If the outcome of 4 holds, and 3 is not yet disproven, you may continue with 341, and so forth; but if the outcome of 4 shows 3 to be false, you will have to go back to 2 and try to invent a new 2, deduce a new 3, look for 4, and so forth.

Note that this method can never absolutely verify (prove the truth of) 2. It can only falsify 2.[3] (This is what Einstein meant when he said, “No amount of experimentation can ever prove me right; a single experiment can prove me wrong.”[4])

## Discussion

Additionally, as pointed out by Carl Hempel (1905–1997), this simple view of the scientific method is incomplete; a conjecture can also incorporate probabilities, e.g., the drug is effective about 70% of the time.[5] Tests, in this case, must be repeated to substantiate the conjecture (in particular, the probabilities). In this and other cases, we can quantify a probability for our confidence in the conjecture itself and then apply a Bayesian analysis, with each experimental result shifting the probability either up or down. Bayes’ theorem shows that the probability will never reach exactly 0 or 100% (no absolute certainty in either direction), but it can still get very close to either extreme. See also confirmation holism.

Qualification of corroborating evidence is sometimes raised as philosophically problematic. The raven paradox is a famous example. The hypothesis that ‘all ravens are black’ would appear to be corroborated by observations of only black ravens. However, ‘all ravens are black’ is logically equivalent to ‘all non-black things are non-ravens’ (this is the contrapositive form of the original implication). ‘This is a green tree’ is an observation of a non-black thing that is a non-raven and therefore corroborates ‘all non-black things are non-ravens’. It appears to follow that the observation ‘this is a green tree’ is corroborating evidence for the hypothesis ‘all ravens are black’. Attempted resolutions may distinguish:

• non-falsifying observations as to strong, moderate, or weak corroborations
• investigations that do or do not provide a potentially falsifying test of the hypothesis.[6]

Evidence contrary to a hypothesis is itself philosophically problematic. Such evidence is called a falsification of the hypothesis. However, under the theory of confirmation holism it is always possible to save a given hypothesis from falsification. This is so because any falsifying observation is embedded in a theoretical background, which can be modified in order to save the hypothesis. Karl Popper acknowledged this but maintained that a critical approach respecting methodological rules that avoided such immunizing stratagems is conducive to the progress of science.[7]

Physicist Sean Carroll claims the model ignores underdetermination.[8]

The hypothetico-deductive model (or approach) versus other research models

The hypothetico-deductive approach contrasts with other research models such as the inductive approach or grounded theory. In the data percolation methodology, the hypothetico-deductive approach is included in a paradigm of pragmatism by which four types of relations between the variables can exist: descriptive, of influence, longitudinal or causal. The variables are classified in two groups, structural and functional, a classification that drives the formulation of hypotheses and the statistical tests to be performed on the data so as to increase the efficiency of the research.