12/3/25, 8:24 AM myMinerva

Academics > HC Resources > HC Handbook > #testability

HC HANDBOOK

#testability
Evaluate whether hypotheses lead to testable
predictions.
Please fill out this Google Form to submit suggestions for this page (i.e.
typos, study guide contributions, useful applications, etc.). We are
actively seeking feedback on anything big or small to make the HC
Handbook as useful as possible.

Defining the HC
Pitch
If it can’t be tested, it falls outside of science.

Paragraph Description
Scientific thinking requires the ability to identify hypotheses that generate
testable predictions. One must learn to identify hypotheses that do not lead to
testable predictions and hence cannot be evaluated. One should also be able to
generate multiple predictions from a hypothesis and determine how well they
will allow us to conclude whether a hypothesis is supported or refuted.

Categorization
Habit | Empirical Analysis | Thinking Critically | Evaluating Claims

[Link] 1/12
12/3/25, 8:24 AM myMinerva

Cornerstone Introduction
Class: Empirical Analysis | Semester One

Unit: Scientific Method

Big Question: How do we mitigate human-caused climate change?

General Example
You work on a local politician's campaign team for the upcoming local elections.
You suggest employing more door-to-door campaigning will increase support
among residents because they seek more face-to-face interactions. Your
colleague argues that while your proposal makes sense, you don't have the means
to determine the impact of door-to-door visits. Your colleague is suggesting that
your hypothesis that increasing face-to-face interactions will cause an increase in
support among residents is untestable. First, you suggest randomizing the
intervention (door-to-door campaigning) among the city’s neighborhoods. Then,
you inform the other campaign channels to keep the same strategy over the
research period. You predict that there will be a larger increase in support among
residents in the neighborhoods that received the door-to-door intervention
compared to the other campaign channels (control groups). While your
hypothesis is testable in principle, however, you still need to find out whether it
is testable in practice. You ask your manager how much budget you would be
able to allot for this experiment and they give you a sufficient amount to cover all
expenses. You then investigate how long your experiment would take and realize
you would need to have the results done by the end of the campaign. Since the
entire research project would take around 2 months, and the elections are in 4
months time, your hypothesis is determined to be testable in both principle and
practice. To improve the testability of the hypothesis even more, you generate a
second prediction: that polling numbers will increase significantly in the
[Link] 2/12
12/3/25, 8:24 AM myMinerva

neighborhoods with door-to-door campaigning compared to the neighborhoods

without. Both of these predictions are useful while testing your hypothesis.

Footnote: Available methods were used to assess the testability of door-to-

door visits. Once it was determined to be testable in both principle and practice,
multiple predictions were generated that will be used to test the hypothesis.

Is it this HC?
Is #testability the right HC? If the answer to the following questions is yes,
#testability could be useful:

1. Are you generating multiple predictions to test a hypothesis?

2. Are you evaluating the testability of a hypothesis in principle and/or
practice?

Depending on the work product, there might be other HCs to consider. ….

The following HC might apply if the

work:

#hypothesisdevelopment Identifies a knowledge gap and generates

an informed conjecture in an attempt to
eliminate the gap. The hypothesis includes
an explanation for a relationship between
variables which should be used to generate
predictions.

#significance Uses different statistical tests to assess the

statistical and practical significance of a

[Link] 3/12
12/3/25, 8:24 AM myMinerva

pattern or relationship, which typically

involves constructing null and alternative
“hypotheses” that serve as predictions to
be tested. Note that because the null and
alternative are two sides of the same
prediction, it counts as one prediction.

#plausibility Evaluates the hypothesis by looking at the

premises and assumptions it is based on.
This HC is typically used along with
#testability to evaluate hypotheses.

#interventionalstudy, Involve designing a study to test a

#observationalstudy, hypothesis. The processes of generating
#casestudy predictions and designing studies are
often iterative, whereby predictions help
determine relevant variables.

#deduction Uses deductive reasoning to derive

conclusions that follow necessarily from a
set of premises, which can be used to
derive testable predictions from a
hypothesis and identify predictions that
can be used to refute a hypothesis.

Self Study
Try answering the questions on your own before checking the example answers.
[Link] 4/12
12/3/25, 8:24 AM myMinerva

👆 What is the difference between testability in principle and

testability in practice? Provide examples to support your
explanation.

Answer:
A hypothesis that is testable in principle is one that leads to predictions that
can either support or refute the hypothesis, whereas a hypothesis that is
testable in practice is one that leads to predictions for which data can
currently be collected to either support or refute the hypothesis. Sometimes a
hypothesis or theory has predictions that could test the hypothesis, but
currently are not possible to test due to technological, ethical, or other
constraints, which would make the hypothesis testable in principle, but not
testable in practice.

For example, Einstein’s Theory of General Relativity made predictions about

how light would be bent in an eclipse, but when published in 1915 this was
only testable in principle because there were no total eclipses on Earth that
could be measured. When a total solar eclipse occurred in 1919, then General
Relativity became testable in practice, because data could be collected that
might support or refute General Relativity. Likewise, General Relativity
predicted a phenomenon of gravitational waves that technology was not able
to measure in 1915. One hundred years later in 2015, technological advances
enabled the first direct measurements of gravitational waves, making that
prediction testable in practice when it previously had been only testable in
principle.

Note that testability in principle always comes first. In other words, a

hypothesis cannot be testable in practice if it is not testable in principle.

[Link] 5/12
12/3/25, 8:24 AM myMinerva

👆 What is the difference between generating multiple

predictions and simply negating an initial prediction?

Answer:
Multiple predictions are different approaches to testing the hypothesis. For
example, a hypothesis that all the continents were arranged together in a
supercontinent (Pangea) in the past was tested using multiple predictions.
One prediction could be: if the supercontinent used to exist, then we would
expect the edges of the current continents to fit together. Another prediction:
if the supercontinent used to exist, we would expect land animal fossils to
match the sides of continents that used to be connected together. A third
could be: if the supercontinent used to exist, rock formations should line up
on the edges of continents that used to be connected together. Although the
predictions are widely different, they all test the same hypothesis. Refer to the
reading on Pangea if you want to learn about other evidence.

An example of a negation of a prediction is the following: the hypothesis

would be supported if we notice a positive correlation between X and Y.
However, if we obtain a negative or no correlation then the results will
contradict the hypothesis. Although it is useful to provide clear
interpretations of different results, the negated prediction does not make the
argument stronger. Note that the negation is not a new prediction.

This is relevant in the realm of #significance when using hypothesis tests to

test the statistical and practical significance of a relationship or pattern. The
alternative hypothesis expresses a prediction about a certain relationship, for
example, there is a positive correlation between X and Y. The null hypothesis
simply represents the status quo, for example, there’s no correlation between

[Link] 6/12
12/3/25, 8:24 AM myMinerva

X and Y. Overall, the significance test is being used to test a single prediction
(e.g., about the existence of a correlation between two variables).

👆 How do we generate multiple predictions?

Answer: There is no single recipe for generating multiple predictions as it

depends on the hypothesis. One way to generate predictions is to vary the
object. For example, if you want to test the effectiveness of an antibiotic your
predictions can be about its effects on cells in petri dishes, and on humans.
You can also think about different relevant variables. For example, one of the
explanations for why men cite themselves more often than women do can be
that men publish more papers, therefore, they have more work to cite. The
first prediction that can test the hypothesis is that there is a positive
relationship between the number of papers published and self-citation in
both, men and women. The second prediction would be that over the first X
papers, the rate of self-citation is similar among men and women. You can see
more examples of multiple predictions in the Minerva University whitepaper
linked in the Key Resources.

👆 How do scientists evaluate competing explanations and how

can #testability aid the generation of effective explanations?

Answer: The main step of evaluating competing hypotheses involves

analyzing how predictions match the observations. The stronger hypothesis
will be the one that more closely explains the observed data. One can use

[Link] 7/12
12/3/25, 8:24 AM myMinerva

#testability to cleverly identify predictions that will enable competing

explanations to be ruled out (i.e., the prediction would clearly support one
hypothesis and refute another).

👆 Explain how #hypothesisdevelopment, #testability, and

#plausibility work together in the process of generating a
hypothesis.

Answer: The first step of generating a hypothesis is to identify patterns of

regularity among observations and suggest an explanation of the pattern
(#hypothesisdevelopment). One needs to ensure that the generated
hypothesis is based on well-established theories and plausible assumptions
(#plausibility). We also need to identify predictions that would either support
or refute the hypothesis and ensure that these predictions are testable given
the constraints (i.e. evaluate whether the hypothesis is testable in principle
and in practice as part of #testability). If the hypothesis is either based on
implausible assumptions or is untestable, we need to go back and revise the
hypothesis.

Applying the HC
Guided Reflection
Have you generated multiple, independent predictions that would clearly
refute or support the hypothesis?
Have you sufficiently explained and justified your process for generating or
deriving predictions?

[Link] 8/12
12/3/25, 8:24 AM myMinerva

Have you identified a prediction that can distinguish between competing

hypotheses?
Have you justified why you focus on a specific prediction?
Have you considered the practicality of testing the prediction (e.g., budget,
duration, ethics)?
Is your interpretation of the expected results insightful for the purpose of
your work?

Common Pitfalls
Provides only a single prediction.
Considers the simple negation of a prediction as an additional prediction.
Provides multiple predictions but they are not connected to the hypothesis
or are not specific enough.
The testability in principle and/or practice is not assessed where needed.
Fails to identify and critique untestable hypotheses.

Applying the HC at Minerva

You can use this HC in your business classes to suggest a test of a
hypothesis regarding the success of a product, or marketing strategy. Very
often, a question will arise regarding the best way to market a product. In
this way, you might hypothesize that the best marketing strategy will be
social media ads for one month as compared to billboards because social
media has the greatest reach. This then leads to clear predictions: (1) a
comparison of sales before and after the social media ad campaign should
show a statistically significant increase in sales post-campaign. (2) In areas
without the social media advertising, sales do not change. (3) One month of
only social media ads should lead to greater sales compared to one month
of only billboards in regions that are similar demographically. Of course,
we should be careful to consider confounding variables!

[Link] 9/12
12/3/25, 8:24 AM myMinerva

This HC is applicable to your research proposal assignments in the Natural

Sciences college. For example, your research proposal on solar
geoengineering technologies would require a thorough analysis of the
testability of your hypothesis given the costly, under-researched, and
ethically controversial nature of such technologies.
You can use #testability to generate and evaluate hypotheses in politics,
economics, and other social sciences. Let’s say you want to conduct
research to test whether the increase in government investment in public
education leads to a decrease in youth unemployment. If the relationship
between the variables (investment in education and youth unemployment)
is true you would expect (1) states with higher government investment to
have higher employment rates of recent graduates, (2) skills provided by
improved public education to match with the demands of the job market
and (3) long term economic growth (reflected in GDP) due to lower
unemployment and high skilled labor. Once you generate multiple
predictions, you may test one or more depending on the time, financial,
and data constraints.

Applying the HC Outside of Minerva

The HC can help you assess claims encountered in your daily
conversations. Imagine a scenario where your neighbor called to warn you
about an unfortunate event foreseen in their dream. Intrigued by the claim,
you decided to collect data on all the negative events throughout your day;
however, nothing particularly bad happened. When you reported this to
your neighbor, they responded by claiming that the future changed because
of their warning. It seems that any outcome, whether the predicted event
occurred or not, could be interpreted as supporting the claim. Furthermore,
even if something negative had happened, the prediction was too broad,
meaning that a random event could wrongly be ascribed as a supporting

[Link] 10/12
12/3/25, 8:24 AM myMinerva

outcome. The inability to falsify makes your neighbor’s claim untestable in

principle.
If your career is in R&D, you will often use #testability to design
experiments by selecting to test the prediction that not only supports a
hypothesis but also contradicts the alternative ones, making research less
costly and time-consuming.

Key Resources
Minerva University. (2020). Developing hypotheses and determining their
testability and plausibility. [Link]
[Link]/uploaded_files/mke/00229870-3142/mu-ns50-
[Link]
This guide outlines the relationship between
#hypothesisdevelopment, #plausibility, and #testability using
multiple examples.

The University of California Museum of Paleontology. (2015). Testing

scientific ideas. Understanding Science.
[Link]
works/testing-scientific-ideas/
This short page explains the importance of testability, and how
predictions and results are used to make judgments about the
hypothesis.

BBC Radio 4. (2015, August 5). Karl Popper’s Falsification. [Video].

Youtube. [Link]
This 2-minute video provides a brief background on Karl Popper’s
ideas on falsification in science. This is useful when considering

[Link] 11/12
12/3/25, 8:24 AM myMinerva

whether a hypothesis is testable in principle, but is not the only way

to design hypotheses.

[Link] 12/12