How to perform descriptive statistical analytics


Would it be good to summarize your data  using a few numbers and graphs?
That's what I'll be talking about in this article. We'll be using descriptive statistics to describe our data.

What do I mean by describing data?
Here's what I mean:

  •  Understanding measures of central tendency for our data.
  • Measure dispersion of our data.
  • Using frequency tables and scatter plots to gain insight.


Understanding Measures of central tendency:

Remember studying about mean, median, and mode in high school?
We use these measures to summarize our data with a single value.
The idea is to find a number that best represent all the numbers we have in a data set.
Let's discuss each one of these measures for better understanding.

Mean : Sum of all values / Number of values 

             example : data = 1,3,5,4,2

                              mean = (1+3 + 5 + 4 +2) / 5 = 3

Median : Arranging all the values in ascending order, and then finding the value in the middle.

            example : data = 1,3,5,4,2
                              arranging data in ascending order : 1,2,3,4,5
                               There are 5 values, thus we pick the 3rd number as median.
                               Median = 3

                   Note  :   Given n is the number of elements in the data set.
                                 pick (n+1)/2th number when n is odd.
                                 pick n/2th number when n is even.      
               
Mode : The value that occurs most frequently in the data.

              example : data = 1,3,5,4,2,1
                                Mode = 1

Note : We prefer using median over mean when there are outliers in the data.
Let me make this point using an example.

                data = 1,2,3,4,5,6,7,8,9,1000
                mean = (1+2+3+4+5+6+7+8+9+1000) / 10 = 104.5
                median = 5
 

The median represents the data better than the mean. The mean gets influenced by one value (1000) in the data. So it is better to use median when there are outliers in the data.

Understanding measures of dispersion :

Want to know about the distribution of our data?
There are two main measures used for this purpose. Range and standard deviation.

Range : Greatest value - Smallest value

example : data = 1,2,3,4,5
                range = 5-1 = 4

 Standard deviation : It tells us what percentage of data lie in a particular rage.
 i.e. : 68% of the data exists above or below the 1st standard deviation.
         95.2% of the data exists above or below the 2nd standard deviation.
         99.2% of the data exists above or below the 3rd standard deviation.



Any value that is outside three standard deviations is considered to be an outlier.

Note : An outlier is a number that is so removed from the norm that is disrupts the data.

Z-Score : Used for comparing a specific value to a population.
 it tells us how standard deviations away from the mean is the value.

Z-score = (Value -  Mean) / Standard deviation

 

Frequency tables and scatter plots :

We want to get an idea of counts of values in specific categories. This is where frequency tables comes in.

Frequency Table : Visualization of counts of values in specific categories.
This is what a frequency table looks like :



But how to make a frequency table with many categories?
This where you will use a contingency table.

Contingency table : Frequency table with many variables
This is what a contingency table looks like :

Let's move on to scatter plots now.

Scatter plot : Visualization of two numerical variables.
This is what a scatter plot looks like :


Any idea why scatter plots are used?
Scatter plots are used to find how two variables are correlated.

Correlation : Measure of how related two variables are.

Three types of correlation :

1) Positive correlation : Both the variables increase together.
2) Negative correlation : One variable increases while the other decreases.
3) No correlation : The the variables are not related.

We measure correlation using correlation coefficient.

Correlation Coefficient : A number that measures how related two variables are.

Correlation coefficient varies between -1 and 1.
A value greater than 0.7 means high positive correlation.
A value less that -0.7 means low negative correlation.

This was all I had to share about descriptive statistics.
Thank you :)


 


Comments

Post a Comment

Popular posts from this blog

The Big Five (Personality Traits)

 Our personalities can be broadly categorized into four categories and we'll use the acronym OCEAN to elaborate on it. Before getting started let's talk about why we need to understand personalities at the very first place. Well it helps in predicting behavior oh other people, and getting to know yourself better, that's why. Our behavior is influenced by two main factors, personality and situation. Back to the OCEAN acronym. These are the big five personality traits of humans : O : Openness C : Conscientiousness E : Extroversion A : Agreeableness N : Neuroticism Let's talk about the big five in some detail. Openness : This trait is all about how an individual perceives new experience. Someone with high openness is likely to be open minded, enjoying change, future oriented. Someone with low openness is likely to be close minded, prefer things to stay as they are, past     oriented. Conscientiousness : Being dutiful and wishing to do one's work well is being con...
Read more

The OSEMN Framework

 Now here's a framework that's super helpful for data science projects. Before applying the framework, we must set the goals and objectives of our project. Goals should always be smart. S - Specific M - Measurable A - Attainable R - Relevant T - Time bound example : Gain 1000 subscribers for a new service in 5 months. Once the goal is set, we should figure out a key performance indicator (KPI) that will help us monitor our growth with respect to our initial goal. example : New Subscribers per month One the goal and KPI are determined, we can start to look at the OSEMN framework to kick off our project. O - Obtain data S - Scrub data E - Explore data M - Model data N - Interpret data Let's look at this framework in some detail. Obtain data : We have to obtain data to analyze it. In our example we can look at previous sales data to figure things out. data can be of 3 types : First party, second party, and third party. data can come from 5 main formats : txt, xlm, csv, json, ...
Read more

The eighty 20

 The 80/20 rule determines that eighty percent of the result comes from twenty percent of action/result. examples : 1) Eighty percent sales from twenty percent of clients. 2) Eighty percent work done in a company by twenty percent of employees. 3) Eighty percent wealth distributed in twenty percent population Therefore choose your twenty percent wisely. You should know why this blog is so short (it only contains the twenty percent)
Read more