1. Create random data

Create random distributed data:

n <- 5 # Number of observations to be sampled.
mean <- 500 # True population mean  
sd <- 100 # True population standard deviation
x <- rnorm(n = n, mean = mean, sd = sd)

You won’t have the same values. Every time you use rnorm you will get new values (i.e. magic). Check the randomly sampled observations by typing x and Return in the R console.

x # Tada, 5 random values

[1] 443.9524 476.9823 655.8708 507.0508 512.9288

2. Why fake data?

Functions like rnorm allow us to control population parameters. In reality we rarely know the population mean, say, the average time it takes to recover from COVID-19, but we can estimate it from a large enough sample and repeated sampling from the population. These unknown population parameters are conventionally indicated with Greek letters, i.e. \(\mu\) (mu; the population mean) and \(\sigma\) (sigma; the population standard deviation).

Because we don’t know the population parameters in real life, functions like rnorm are great. We can simulate data of which we do know the parameter values, because we defined both above (i.e. mean, sd).

3. Plot the data

Plot the “fake” data from above.

histogram(x, data = NULL)

Here is the mean of the sample:

mean(x)

[1] 519.357

and the standard deviation of the sample:

sd(x)

[1] 81.10218

1. Question

Your numbers will be different from mine because we take random samples. Are the data in your histogram normal distributed? Why do you think they are (or are not) normal distributed?

2. Question

Compare the sample mean and the sample standard deviation to our population mean and population standard deviation. Are they different? Remember, we are in the almost unique situation to know the population parameters. Why do you think are the sample mean and standard deviation different from the population mean and standard deviation?

1. Task

Repeat the steps above but this time create a histogram with sample size n=1000. What changes did you make in the code? How has the histogram changed? Calculate the sample mean and sd again. How did they change?

2. Task

Repeat the steps above but this time with a sample size n=1000 and a sd of 10. What changes did you make in the code? How has the histogram changed? Calculate the sample mean and sd again. How did they change?

3. Task

If we know the population mean and population variance, we know everything we need to know to create data we would be expected under the normal distribution. Let’s do this for IQ which has a population mean of 100 and a population sd of 15. Use an n of 1000 samples and plot a histogram of IQ.

1. Task

Calculate the probability of observing a person with an IQ below 75 corresponding to the area shaded in this plot:

2. Task

Calculate the probability of observing a person with an IQ above 75 corresponding to the area shaded in the plot below.

Hint: Keep in mind that pnorm returns the probability of IQ lower than the critical value. The entire area underneath the curve sums to 1 (100%) and the distribution is symmetric. In other words you can use 1 - x where x is the result returned from pnorm. Make sure you understand why :)

3. Task

Calculate the probability of observing a person with an IQ between 95 and 110 corresponding to the area shaded in the plot below. Example below.

As an example, here is the probability of observing a person with an IQ between 75 and 95

pnorm(95, mean = mean, sd = sd) - pnorm(75, mean = mean, sd = sd)

[1] 0.321651

which is the probability of observing a person with an IQ below 95

pnorm(95, mean = mean, sd = sd)

[1] 0.3694413

but removing the probability of observing a person with an IQ below 75

pnorm(75, mean = mean, sd = sd)

[1] 0.04779035

4. Task

Calculate the probability of observing a person with an IQ between 99.9 and 100.1 corresponding to the area shaded in the plot below.

5. Task

Calculate the probability of observing a person with an IQ of below 60 or above 140; see the area shaded in the plot.

6. Task (bonus)

Calculate the probability of observing a person with an IQ outside of one standard deviation around the mean (see plot). Reminder the population mean of IQ is 100 and its standard deviation is 15.

1. Question

What type of data are the responses to each item?

2. Question

Are responses to these items are normal distributed? Why (or why not)?

1. Task

First we need to define the response options available to the participant.

# response options
response <- 1:5 # people can only respond with a number from 1 to 5
# Check out the vector
response # numbers from 1 to 5 (easy)

[1] 1 2 3 4 5

We can use sample to take random samples from response. Try out; there is a \(\frac{1}{5}\) chance the number you obtained is the same as mine. Do you know why?

sample(response, size = 1) # Sample 1 random number between 1 and 5

[1] 4

What’s the code for sampling 2 random numbers?

2. Task

To sample more than 5 random numbers we need to set replace = TRUE to allow the same number to be sample more than once (called sampling with replacement).

sample(response, size = 6, replace = TRUE) # Sample 6 random numbers between 1 and 5

[1] 1 2 3 5 3 3

Remember that the depression scale above has 22 items. So if we want to simulate one participant who responds to every of the 22 items (at random), what would you need to change size to?

3. Task

You probably worked out in Task 3 that size has to be 22. Lets save the result in ppt_1.

ppt_1 <- sample(response, size = 22, replace = TRUE) # Simulate one participant

Let’s plot these data to see how they are distributed. Use histogram() and set data to NULL (cause we don’t use a data frame) and set x to ppt_1. Replace xxxs.

# create histogram of ppt_1
#histogram(data = NULL, x = xxx) 

Is this distribution normal? Why or why not?

4. Task

We’ve seen how we can sample random data for one participant that answers all 22 depression items. Now, lets demonstrated the central limit theorem. Remember that the central limit theorem states that we will arrive at a normal distribution if our sample is approaching infinity (or a large number) regardless of the shape of the distribution we’re sampling from.

We don’t learn much from sampling data for one hypothetical participant. We can use the replicate function to do the same for 2 participants.

replicate(2, sample(response, size = 22, replace = T))

      [,1] [,2]
 [1,]    5    1
 [2,]    5    2
 [3,]    3    4
 [4,]    1    5
 [5,]    2    5
 [6,]    5    3
 [7,]    5    1
 [8,]    4    4
 [9,]    5    1
[10,]    2    1
[11,]    1    3
[12,]    1    4
[13,]    3    1
[14,]    1    3
[15,]    5    5
[16,]    1    3
[17,]    2    2
[18,]    4    5
[19,]    4    5
[20,]    3    3
[21,]    1    2
[22,]    2    2

Let’s save the data for two participants in two_ppts and calculate the means for each column (i.e. participant) using colMeans.

two_ppts <- replicate(2, sample(response, size = 22, replace = T))
two_ppts_means <- colMeans(two_ppts)

These are the means:

two_ppts_means

[1] 3.227273 2.909091

Here is a histogram of two participant means:

histogram(data = NULL, x = two_ppts_means)

This was good but still not very impressive in terms of normal distributions. Get a histogram for 10, 100 and then 1000 hypothetical participants. Replace xxx accordingly and create a histogram each time.

# samps <- replicate(xxx, sample(response, size = 22, replace = T))

#samps_means <- colMeans(xxx)

#histogram(data = NULL, x = xxx) 

Share the plot that you think shows a normal distribution best in your Teams chat:

To share your plot: Click on Export in the Plots panel, then Copy to Clipboard ... and Copy Plot, then go to your Teams chat, click in the chat box and click CTRL + V to insert the plot (or CMD + V on Mac).

5. Task (bonus)

You feel the previous task wasn’t challenging enough? Try to do the same histogram of the sampling distribution again but instead of means, use total; i.e. replace colMeans with colSums. The central tendency theorem works not just for means but also for totals and even for standard deviations.