John Tukey (1962)

John Tukey pioneered Exploratory Data Analysis starting in 1962 and again with a book in 1977.

Summary Statistics are focused on the location, variability, and distribution of data.

Let’s imagine a variable showing the heights of different dogs.

The dog’s heights (in mm) are: 600, 470, 170, 430, and 300

Mean is the sum of all values divided by the number of values.

AKA “average”

\(\dfrac{600+470+170+430+300}{5} = 394\)

We can calculate the mean easily in Python.

# Put the dog heights into a dataframe
dogs = pd.DataFrame({"height": [600, 470, 170, 430, 300]})

dogs.height.mean() # Calculate the mean

Median is the value such that half of the data lies above and below.

AKA “50th percentile”

600, 470, 170, 430, 300

dogs.height.median()

Percentile is a value such that P percent of the data lies below.

AKA “quantile”

The 25th Percentile is the 1st Quartile.

600, 470, 170, 430, 300

dogs.describe()

The 75th Percentile is the 3rd Quartile.

600, 470, 170, 430, 300

dogs.describe()

The median is the 2nd Quartile!!

An outlier is a data value that’s different from most of the data.

AKA “extreme value”

A robust variable is not sensitive to extreme values.

The interquartile range is the difference between the 1st and 3rd quartiles.

dogs.height.quantile(.75)-dogs.height.quantile(.25)

A deviation is the difference between an actual value and an estimate of location (like the mean).

The variance is the sum of the squared deviations, divided by the number of values.

\(\dfrac{206^2+76^2+(-224)^2+36^2+(-94)^2}{5} = 21,704\)

The standard deviation is the square root of the variance.

It gives us a “standard” way of knowing what is normal, or what is extra large/extra small.

• Rottweilers are tall, and dachsunds are short—compared to the standard deviation from the mean.
• Outliers are often more than two standard deviations from the mean.

Now calculate the variance and standard deviations in Python.

dogs.height.var()

dogs.height.std()

Were these the results you expected?

Population vs. Sample

• Population: the full data that exists (i.e. all the dogs in the world)
• Sample: the actual data that was collected (i.e. our set of 5 dogs)

Population vs. Sample

When you have “N” data values:

• The Entire Population: divide by N when calculating variance (like we did)

• A Sample: divide by N-1 when calculating variance

• Sample variance: \(\dfrac{108,520}{4}=27,130\)
  

• Sample standard deviation: \(\sqrt{27,130}=164\)

• Think of it as a “correction” when your data is only a sample. Pandas does this by default!

Neither the mean, variance, nor standard deviation are robust. They are all very sensitive to outliers!

Histograms show distributions based on frequency counts.

The histogram for miles per gallon highway in the mpg dataset.

The normal distribution has most values in the middle.

In a normal distribution, 95% of the values lie within 2 standard deviations of the mean.

Be careful: normal distributions are assumed for many statistical analyses!

Boxplots show distribution based on the median.

You can see how the box plot and the histogram are similar but different.

Location, Variability, and Distribution are for one variable at a time (univariate analysis). Correlation is for two variables (bivariate analysis).

Scatter plots and correlation coefficients are used to determine correlation.

We’ll talk more about correlation in a couple weeks!

Resampling is simply drawing multiple random samples from observed data.

I can grab a sample of 5 observations. Then I can “resample” 5 more. Then 5 more, and so on and so on.

First proposed in the 1960s, resampling procedures weren’t practical until computing took off in the 1980s.

Resampling is an umbrella term. It can include:

• The Bootstrap, used to assess the reliability of an estimated statistic
• Permutation Tests, used as an alternative to parametric hypothesis tests

You can resample with or without replacement.

Replacement means an item is returned to the sample before the next draw (i.e. you could wind up with the same observation multiple times).

Using bootstrap sampling, we can estimate a confidence interval for any summary statistic.

• Remember: bootstrap sampling is sampling with replacement.
• The 90% confidence interval tells us where 90% of the data is likely to fall, if the data were random.
• Confidence intervals are used to gauge uncertainty.

Confidence intervals with Pandas

Work with a partner to calculate the confidence interval for the mean height in the dogs dataframe (remember to use your Pandas cheatsheet and past slideshows):

Confidence intervals with Pandas

n_rows = dogs.shape[0] # First find the number of rows in the dataframe.

# Then take a sample with replacement from your dataset, and
# Calculate the mean each time. Do this 5000 times.
bootstrap_samples = []
for i in range(5000):
    sample_mean = dogs.sample(n=n_rows, replace=True).height.mean()
    bootstrap_samples.append(sample_mean)

# Put the results into a pandas Series object
bootstrap_samples = pd.Series(bootstrap_samples)

# Calculate the 95th percentile and the 5th percentile.
top_percentile = bootstrap_samples.quantile(.95)
bottom_percentile = bootstrap_samples.quantile(.05)

# Print the results using nice f-strings.
print(f"The mean dog height in our data is {dogs.height.mean():.3f}, with a 90% confidence interval of {bottom_percentile:.3f} to {top_percentile:.3f}.")