Relative error of

point estimates
SAMPLING IN R

Richie Cotton
Data Evangelist at DataCamp
Sample is number of rows
coffee_ratings %>% coffee_ratings %>%
slice_sample(n = 300) %>% slice_sample(prop = 0.25) %>%
nrow() nrow()

300 334

SAMPLING IN R
Various sample sizes
coffee_ratings %>% coffee_ratings %>%
summarize(mean_points = mean(total_cup_points)) %>% slice_sample(n = 10) %>%
pull(mean_points) summarize(mean_points = mean(total_cup_points)) %>%
pull(mean_points)

82.15 82.82

coffee_ratings %>% coffee_ratings %>%

slice_sample(n = 100) %>% slice_sample(n = 1000) %>%
summarize(mean_points = mean(total_cup_points)) %>% summarize(mean_points = mean(total_cup_points)) %>%
pull(mean_points) pull(mean_points)

82.02 82.16

SAMPLING IN R
Relative errors
Population parameter

population_mean <- coffee_ratings %>%

summarize(mean_points = mean(total_cup_points)) %>%
pull(mean_points)

Point estimate

sample_mean <- coffee_ratings %>%

slice_sample(n = sample_size) %>%
summarize(mean_points = mean(total_cup_points)) %>%
pull(mean_points)

Relative error as a percentage

100 * abs(population_mean - sample_mean) / population_mean

SAMPLING IN R
Relative error vs. sample size
ggplot(errors, aes(sample_size, relative_error)) +
geom_line() +
geom_smooth(method = "loess")

SAMPLING IN R
Let's practice!
SAMPLING IN R
 Creating a sampling
distribution
SAMPLING IN R

Richie Cotton
Data Evangelist at DataCamp
Same code, different answer
coffee_ratings %>% coffee_ratings %>%
slice_sample(n = 30) %>% slice_sample(n = 30) %>%
summarize(mean_cup_points = mean(total_cup_points)) %>% summarize(mean_cup_points = mean(total_cup_points)) %>%
pull(mean_cup_points) pull(mean_cup_points)

83.33 82.59

coffee_ratings %>% coffee_ratings %>%

slice_sample(n = 30) %>% slice_sample(n = 30) %>%
summarize(mean_cup_points = mean(total_cup_points)) %>% summarize(mean_cup_points = mean(total_cup_points)) %>%
pull(mean_cup_points) pull(mean_cup_points)

82.16 82.25

SAMPLING IN R
Same code, 1000 times
mean_cup_points_1000 <- replicate( [1] 81.65 81.57 82.66 82.27 81.76 81.74 82.71
n = 1000, [8] 82.20 80.43 82.45 82.29 82.63 82.28 82.11
expr = coffee_ratings %>% [15] 82.14 81.72 81.97 82.58 81.78 82.47 81.73
slice_sample(n = 30) %>% [22] 82.78 82.14 82.39 81.69 82.36 82.64 82.68
summarize( [29] 82.56 82.14 82.72 82.43 81.68 82.74 82.80
mean_cup_points = mean(total_cup_points) [36] 82.12 82.31 81.02 82.83 81.71 82.25 82.11
) %>% [43] 82.76 82.26 81.57 82.00 81.75 81.47 81.99
pull(mean_cup_points) [50] 82.68 82.05 82.43 82.40 82.66 80.78 82.43
) ...
[967] 81.84 83.12 81.54 81.83 82.24 82.36 82.49
[974] 82.05 82.08 81.98 82.45 82.04 81.42 83.06
[981] 81.97 82.65 81.12 82.48 81.64 81.92 81.96
[988] 81.71 81.96 81.78 82.30 81.76 82.46 82.43
[995] 81.95 82.60 81.84 82.78 82.23 82.56

SAMPLING IN R
Preparing for plotting
library(tibble) # A tibble: 1,000 x 1
sample_means <- tibble( sample_mean
sample_mean = mean_cup_points_1000 <dbl>
) 1 83.3
2 82.6
3 82.2
4 82.2
5 81.7
6 81.6
7 82.7
8 82.3
9 81.8
10 81.7
# ... with 990 more rows

SAMPLING IN R
Distribution of sample means for size 30
ggplot(sample_means, aes(sample_mean)) +
geom_histogram(binwidth = 0.1)

A sampling distribution is a distribution of

several replicates of point estimates.

SAMPLING IN R
Different sample sizes
Sample size 6 Sample size 150

SAMPLING IN R
Let's practice!
SAMPLING IN R
 Approximate
sampling
distributions
SAMPLING IN R

Richie Cotton
Data Evangelist at DataCamp
4 dice
# A tibble: 1,296 x 4
die1 die2 die3 die4
<int> <int> <int> <int>
1 1 1 1 1
2 1 1 1 2
3 1 1 1 3
library(tidyr) 4 1 1 1 4
dice <- expand_grid( 5 1 1 1 5
die1 = 1:6, 6 1 1 1 6
die2 = 1:6, 7 1 1 2 1
die3 = 1:6, 8 1 1 2 2
die4 = 1:6 9 1 1 2 3
) 10 1 1 2 4
# ... with 1,286 more rows

SAMPLING IN R
Mean roll
dice <- expand_grid( # A tibble: 1,296 x 5
die1 = 1:6, die1 die2 die3 die4 mean_roll
die2 = 1:6, <int> <int> <int> <int> <dbl>
die3 = 1:6, 1 1 1 1 1 1
die4 = 1:6 2 1 1 1 2 1.25
) %>% 3 1 1 1 3 1.5
mutate( 4 1 1 1 4 1.75
mean_roll = (die1 + die2 + die3 + die4) / 4 5 1 1 1 5 2
) 6 1 1 1 6 2.25
7 1 1 2 1 1.25
8 1 1 2 2 1.5
9 1 1 2 3 1.75
10 1 1 2 4 2
# ... with 1,286 more rows

SAMPLING IN R
Exact sampling distribution
ggplot(dice, aes(factor(mean_roll))) +
geom_bar()

SAMPLING IN R
The number of outcomes increases fast
outcomes <- tibble(
n_dice = 1:100,
n_outcomes = 6 ^ n_dice
)

ggplot(outcomes, aes(n_dice, n_outcomes)) +

geom_point()

SAMPLING IN R
Simulating the mean of four dice rolls

four_rolls <- sample(

1:6, size = 4, replace = TRUE
)
mean(four_rolls)

SAMPLING IN R
Simulating the mean of four dice rolls
sample_means_1000 <- replicate( # A tibble: 1,000 x 1
n = 1000, sample_mean
expr = { <dbl>
four_rolls <- sample( 1 4
1:6, size = 4, replace = TRUE 2 4.5
) 3 2.5
mean(four_rolls) 4 3.75
} 5 3.75
) 6 4
7 3

sample_means <- tibble( 8 4.75

sample_mean = sample_means_1000 9 3.75

) 10 4.25
# ... with 990 more rows

SAMPLING IN R
Approximate sampling distribution
ggplot(sample_means, aes(factor(sample_mean))) +
geom_bar()

SAMPLING IN R
Let's practice!
SAMPLING IN R
 Standard errors and
the Central Limit
Theorem
SAMPLING IN R

Richie Cotton
Data Evangelist at DataCamp
Sampling distribution of mean cup points

SAMPLING IN R
Consequences of the central limit theorem
Averages of independent samples have approximately normal distributions.
As the sample size increases,

the distribution of the averages gets closer to being normally distributed, and
the width of the sampling distribution gets narrower.

SAMPLING IN R
Population & sampling distribution means
coffee_ratings %>% Sample size Mean sample mean
summarize(
5 82.1496
mean_cup_points = mean(total_cup_points)
) %>% 20 82.1610
pull(mean_cup_points)
80 82.1496

82.1512 320 82.1521

SAMPLING IN R
Population & sampling distribution standard deviations
coffee_ratings %>% Sample size Std dev sample mean
summarize( 5 1.1929
sd_cup_points = sd(total_cup_points)
20 0.6028
) %>%
pull(sd_cup_points) 80 0.2865

320 0.1304
2.68686

SAMPLING IN R
Population mean over square root sample size
Sample size Std dev sample mean Calculation Result
5 1.1929 2.68686 / sqrt(5) 1.2016

20 0.6028 2.68686 / sqrt(20) 0.6008

80 0.2865 2.68686 / sqrt(80) 0.3004

320 0.1304 2.68686 / sqrt(320) 0.1502

SAMPLING IN R
Let's practice!
SAMPLING IN R