Causation vs Association, and an Introduction to Experiments (3.1)

Summary

This video script delves into the distinction between observational studies and experiments in scientific research. It emphasizes that while correlation is evident in observational studies, such as the link between study time and GPA, it doesn't imply causation due to lurking variables like IQ and motivation. To establish causation, experiments are necessary, involving the application of treatments to experimental units to observe effects. The script uses an example of an experiment on oral health, explaining key terms like response variable, experimental unit, factors, and factor levels. It concludes by highlighting the principles of randomization, repetition, and control in experiments, which provide stronger evidence for causation than mere observation.

Takeaways

• 📊 Observational studies only measure variables of interest without manipulating them, like studying the correlation between study time and GPA.
• 🔍 Correlation does not imply causation; for example, more study time correlates with higher GPA, but it doesn't mean studying more causes a higher GPA due to lurking variables like IQ and motivation.
• 🔬 To avoid the influence of lurking variables, experiments are performed where a treatment is applied to an experimental unit to provoke a measurable response.
• 🧪 Experiments are distinct from observational studies by actively doing something to the experimental unit rather than just observing what naturally occurs.
• 💊 The experimental unit is the subject of the experiment, which can be almost anything, such as a sick cat in the example of choosing medicine.
• 🧬 A factor is the explanatory variable in an experiment that causes change, like the type of medicine in the cat's treatment.
• 💊 Factor levels are the specific conditions of a factor, such as different brands of medicine like Advil and Tylenol.
• 🤝 A treatment is the combination of factor levels applied to an experimental unit, which can be different combinations of brushing time and toothpaste brand in the oral health experiment.
• 👥 If the experimental units are human, they are called subjects, and in the toothpaste example, healthy individuals are the subjects.
• 📈 The number of treatments in an experiment can be calculated by multiplying the factor levels together, like 2 brushing times and 3 toothpaste brands yielding 6 treatments.
• 🔢 For each treatment, a certain number of experimental units are needed, calculated by multiplying the number of units per treatment by the total number of treatments, like 5 individuals per treatment times 6 treatments equals 30 individuals.
• 📝 A proper experiment follows principles of randomization, repetition, and control to ensure unbiased results and evidence for causation.

Q & A

What is the main difference between observational studies and experiments?

Observational studies involve only measuring variables of interest without intervening, while experiments involve applying a treatment to an experimental unit to observe its effects, allowing for the investigation of causation.

Why does correlation in observational studies not necessarily imply causation?

Correlation does not imply causation because there may be lurking variables that affect the relationship between the explanatory variable and the response variable, which are not accounted for in observational studies.

What are lurking variables and how do they affect the results of observational studies?

Lurking variables are hidden factors that can influence the relationship between the explanatory and response variables. They can affect the results of observational studies by confounding the observed correlation, making it difficult to establish causation.

What is an experimental unit in the context of an experiment?

An experimental unit refers to the entity on which the experiment is being performed, which could be almost anything depending on the nature of the experiment.

What is the difference between a treatment and a factor in an experiment?

A treatment is the experimental condition being applied to an experimental unit, which is a combination of factor levels. A factor is the explanatory variable of an experiment, representing what causes the change being studied.

What are factor levels in an experiment?

Factor levels are the specific conditions or settings of a factor in an experiment, which are applied to the experimental units to observe their effects on the response variable.

How does the example of Dr. Liam's toothpaste experiment illustrate the concept of treatments?

In Dr. Liam's experiment, treatments are the combinations of brushing time (30 seconds and 2 minutes) and toothpaste brand (Colgate, Crest, Sensodyne) applied to healthy individuals to observe their effects on oral health.

What is the purpose of randomization in conducting an experiment?

Randomization is used to randomly allocate experimental units to treatments to prevent biased results and ensure that the experiment's findings are more reliable and generalizable.

What does repetition in an experiment entail and why is it important?

Repetition refers to applying a treatment to multiple experimental units to reduce variation in the results. It is important because it strengthens the evidence for the experiment by demonstrating consistent responses across different units.

What is the role of a control in an experiment?

A control serves as a baseline for comparison against other treatments in an experiment. It helps to isolate the effects of the experimental conditions by providing a reference point to measure changes against.

Why are experiments considered superior to observational studies in establishing causation?

Experiments are superior because they allow for the manipulation of variables and control over conditions, which provides evidence for causation. Observational studies can only show association, not causation, due to the potential influence of lurking variables.

Outlines

00:00

🔍 Understanding Observational Studies and Experiments

This paragraph delves into the distinction between observational studies and experiments in research. It emphasizes that while observational studies involve measuring variables of interest, such as height or test scores, they cannot establish causation due to the potential influence of lurking variables like IQ, motivation, and stress. The paragraph introduces the concept of causation versus association, explaining that correlation does not imply causation, as illustrated by the example of study time and GPA. To overcome the limitations of observational studies, experiments are proposed, which involve applying treatments to experimental units to observe effects. The paragraph also defines key terms such as experimental unit, treatment, factor, and factor levels, using the example of a doctor choosing medicine for a sick cat. It concludes with an example of an experiment conducted by Dr. Liam to study the impact of brushing time and toothpaste brand on oral health, highlighting the importance of defining the response variable, experimental unit, factors, and factor levels.

05:02

📊 Designing an Experiment: Principles and Structure

The second paragraph focuses on the design and execution of an experiment, particularly emphasizing the principles of randomization, repetition, and control. It explains that randomization involves the random allocation of experimental units to treatments to prevent biased results, while repetition refers to applying the same treatment to multiple units to reduce variability in outcomes. The paragraph uses the example of an experiment with 30 healthy individuals, divided into six groups of five, each receiving a different treatment combination. It details the process of calculating the number of treatments and the number of individuals needed for the experiment, based on the factor levels of brushing time and toothpaste brand. The paragraph concludes by discussing the importance of a control group in comparing treatments and the overall benefits of experiments over observational studies in providing evidence for causation.

Keywords

💡Observational Studies

Observational studies are research methods where researchers measure variables of interest without manipulating them. In the video, it is mentioned that these studies only require measuring, such as an individual's height or test scores. The key point made is that correlation or association found in observational studies does not imply causation. For instance, a positive correlation between study time and GPA does not mean that more study time causes a higher GPA due to lurking variables like IQ or motivation.

💡Correlation

Correlation in the context of the video refers to a statistical relationship between two variables, such as the relationship between study time and GPA. It is emphasized that a strong correlation does not necessarily mean that one variable causes the change in the other, which is a critical distinction in understanding the limitations of observational studies.

💡Causation

Causation is the relationship where one event or variable causes another to occur. The video script explains that while observational studies can show correlation, they cannot prove causation. This is contrasted with experimental studies, which aim to establish causation by manipulating variables and observing effects.

💡Lurking Variables

Lurking variables, as mentioned in the script, are hidden factors that can affect the relationship between the explanatory variable and the response variable. They are a potential source of bias in observational studies, as they can influence outcomes without being measured or controlled for, such as the example of IQ affecting both study time and GPA.

💡Experiment

An experiment, as described in the video, is a scientific method where researchers apply a treatment to an experimental unit to observe its effects. Unlike observational studies, experiments involve active manipulation to provoke a response, which is then measured. The video uses the example of a doctor choosing medicine for a sick cat to illustrate the concept of an experiment.

💡Experimental Unit

The experimental unit is the entity on which an experiment is performed. In the video, it is used to refer to the subjects of the experiment, such as the healthy individuals in Dr. Liam's toothpaste study. Each experimental unit receives a specific treatment, and their response is measured to assess the experiment's outcome.

💡Treatment

Treatment in an experiment refers to the experimental condition applied to an experimental unit. The video script clarifies that a treatment is not just a single factor but a combination of factor levels. For example, in the toothpaste study, treatments are combinations of brushing times and toothpaste brands applied to individuals.

💡Factor

A factor in an experiment is the explanatory variable that researchers believe may cause a change in the response variable. The video script uses the example of brushing time and the brand of toothpaste as factors in Dr. Liam's experiment, which are the general categories being tested for their effect on oral health.

💡Factor Levels

Factor levels are the specific conditions or settings of a factor in an experiment. The video provides the example of brushing time having levels of 30 seconds and 2 minutes, and toothpaste brands having levels of Colgate, Crest, and Sensodyne. These levels are combined to form treatments.

💡Randomization

Randomization in the context of experiments is the process of randomly assigning experimental units to different treatments. The video script explains that this principle helps prevent biased results by ensuring that each treatment group is representative of the overall population.

💡Repetition

Repetition in experiments refers to the application of a treatment to multiple experimental units to reduce variation in the results. The video script mentions that having five healthy individuals treated at each combination of factor levels is an example of repetition, which strengthens the evidence provided by the experiment.

💡Control

Control in experiments involves comparing the effects of two or more treatments to account for the influence of other variables. The video script uses the toothpaste study as an example, where the level of oral health after one month serves as a control to compare the effects of different treatments on oral health.

Highlights

Observational studies require only measuring a variable of interest, such as an individual's height or test score.

Correlation or association in observational studies does not imply causation, as seen with study time and GPA.

Lurking variables, like IQ, motivation, and stress, can affect the relationship between explanatory and response variables.

Experiments are conducted to avoid lurking variables by actively applying treatments to provoke responses.

The experimental unit is the subject of the experiment, which can be anything the experiment is performed on.

A treatment in an experiment is the experimental condition applied to an experimental unit.

Factors are the explanatory variables that cause change in an experiment, such as the type of medicine for a sick cat.

Factor levels refer to the specific experimental conditions, like different brands of medicine.

A treatment can be defined as the combination of factor levels applied to an experimental unit.

If experimental units are human, they are called subjects.

Dr. Liam's experiment aims to examine the effectiveness of brushing time and toothpaste brand on oral health.

The response variable in the experiment measures the outcome, which is the level of oral health.

The experimental unit in Dr. Liam's study is a healthy individual.

Factors in the experiment include brushing time and the brand of toothpaste.

Factor levels are specific times for brushing and specific brands of toothpaste.

Treatments are combinations of factor levels, resulting in a total of six treatments in the toothpaste experiment.

Five healthy individuals are needed for each treatment combination, totaling 30 individuals for the experiment.

A proper experiment follows principles of randomization, repetition, and control to ensure validity.

Randomization involves the random allocation of experimental units to treatments to prevent biased results.

Repetition involves applying a treatment to multiple experimental units to reduce variation in results.

Control involves comparing treatments to manage the effects of possible working variables.

Experiments provide evidence for causation, distinguishing them from observational studies.

Transcripts

foreign

causation versus Association and

experiments so far what we have been

looking at were observational studies

observational studies require only

measuring a variable of interest this

includes measuring an individual's

height and recording a person's test

score

when looking at observational studies we

must remember that correlation or

Association does not imply causation for

example we saw a strong positive

correlation with study time and GPA but

that doesn't necessarily mean that

studying more automatically gives you a

better GPA there are some people that

study less than others and still get a

higher GPA this is because there are

other variables to consider like IQ the

amount of motivation stress sleep and

many more these are called lurking

variables because they are hidden

variables that can affect the

relationship between the explanatory

variable and the response variable

to avoid lurking variables an experiment

must be performed this is different from

an observational study in an

observational study you go there only to

observe and you measure and make

associations from things that are

already happening by Nature for an

experiment we are actually trying to

provoke a response to get results by

actually doing something to someone

instead of just observing

an experiment involves applying a

treatment to an experimental unit and

observing their effects this response

can range from either good or bad and

the response is what we are measuring so

there are some definitions you'll have

to know the experimental unit refers to

almost anything that the experiment is

being performed on

a treatment is the experimental

condition being applied to a unit

now if we assume that the cat is sick

the doctor needs to think about what

type of medicine is best for curing the

cat this is known as a factor and it

refers to the explanatory variable of an

experiment they are what causes of the

change

the doctor also has to think about what

brand of medicine will work best because

there are many brands that can do the

same thing

for example we know that Advil and

Tylenol can both help for curing

headaches so the names of brands are

called Factor levels they refer to the

experimental condition being applied to

a unit

notice how we have the same definition

for the treatment and the factor levels

these two definitions are actually

slightly different

now that we know what a factor level is

we can improve the definition of a

treatment the treatment actually refers

to the combination of factor levels

being applied to a unit also if the

experimental units are human then they

are called subjects

these definitions will start to make

more sense after an example so suppose

Dr Liam wants to conduct an experiment

to examine the effectiveness of brushing

time and the brand of toothpaste on oral

health a total of five healthy

individuals will be treated at each

combination of factor levels after one

month the level of oral health will be

compared identify each of the following

in this experiment

this looks like a lot to read but these

are the types of questions you will see

on your exam the response variable

measures the outcome of the study it is

what we are comparing so it would be the

level of oral health

the experimental unit is referring to a

healthy individual it is not just any

type of individual it must be a healthy

individual

now let's look at the factors

you can think of the factor as being the

most General category so we have

brushing time and the brand of

toothpaste these are the factors of the

experiment

Factor levels are more specific than a

factor so instead of just brushing time

we will look at the actual time so it's

talking about the 30 seconds and 2

minutes and instead of just toothpaste

it's talking about the types of

toothpaste so Colgate Crest and

Sensodyne these are our Factor levels

on the other hand a treatment is the

combination of factor levels to

determine the treatments we can make a

table we will have brushing time on one

side of the table and the brand of

toothpaste on the other side of the

table we will write out the factor

levels for each factor for brushing time

we have 30 seconds and 2 minutes for the

brand of toothpaste we have Colgate

Crest and Sensodyne we will then put

these Factor levels together so I will

have 30 seconds with Colgate 30 seconds

with Crest 30 seconds with Sensodyne and

so on these are the treatments that we

are applying to the experimental units

and we see that there is a total of six

treatments

to determine how many treatments you

have you can multiply the factor levels

together so we will have two times three

which is equal to the six treatments

to determine the number of individuals

needed for the experiment we will have

to look back at the original question

it says that a total of five healthy

individuals will be treated at each

combination of factor levels that means

we need to put five people in each

treatment

so we will need a total of 30 healthy

individuals which is equal to the number

of people in each treatment times the

number of treatments

so to recap we can make a diagram of the

experiment we started off with 30

healthy individuals then we randomly

assign these people into six groups of

five

each group will get their own treatment

and at the end we will be comparing the

oral health we can then decide to

calculate correlation and regressions to

help us make any final conclusions about

the results of the experiment

a proper experiment follows three basic

principles

randomization repetition and control

randomization is the random allocation

of experimental units to treatments this

was when we randomly assigned the 30

healthy individuals into different

groups to receive different treatments

randomization helps us to prevent biased

results

repetition refers to a treatment being

done on many experimental units we do

this to reduce the variation in the

results

this refers to the five people in each

group receiving a treatment

we are repeating a treatment by doing it

five times on five different people

generally it's better to have many

experimental units receiving a treatment

than only one because we should expect

that all five experimental units

receiving the treatment to give the same

response this can help us provide

evidence for an experiment

control is the comparison of two or more

treatments to control the effects of

possible working variables in the

toothpaste example it was comparing the

oral health after one month of these

treatments in conclusion experiments are

better than observational studies

because they provide evidence for

causation