Baby Boom: Udemy Excel Tutorial on Analyzing Large Data Sets
This tutorial not only shows how to use Excel Pivot Tables and Graphs, but teaches the mindset needed in exploratory data analysis - look beneath the surface, consider the non-obvious interpretations, and question everything (including the data).
Some of it is addressed to a more beginner level, but here are the parts more interesting to KDnuggets readers and Data Scientists. The data is California baby names from the US Social Security Baby Names Database.
The tutorial uses Excel 2013, so if you have a different version, may see slight differences as you go through the steps.
Download the data and import it into Excel
Download the state-specific data from http://www.ssa.gov/oact/babynames/limits.html. You’ll find a file named namesbystate.zip in your download folder. Extract the California file: CA.TXT. (In Windows, you can just drag the file out of the archive.)
Launch Microsoft Excel, and open CA.TXT. If you don’t see the file in your dialogue box, you may have to choose Show All Files in the dropdown box next to the file name box.
In the Text Import dialogue box, choose Delimited, then Next, then Comma, then Finish. This tells Excel to treat commas as column separators. Save your file as an Excel Workbook file called CA Baby Names.xlsx.
Select the first column, A, and delete it; all of your data in this file is from California, you don’t need to waste computer resources on that information. Insert a new row above Row 1, and type column headers: Sex, Year, Name and Births.
When doing data analysis, it’s essential to take a step back every now and then and ask, “do these results make sense?” This is especially important when you are changing the values of cells in an Excel Spreadsheet; if you make a mistake and change your data, it can be difficult to track down the error later.
But sanity checks can also be used to check the state of the data as it came to you. Use the filter to select only the name ‘Jennifer’, and have a look at the results. The following things should stand out:
- On your way down the list, there were quite a few names that were almost, but not quite, spelled ‘Jennifer’, like ‘Jennfier’ and ‘Jenniffer’. Some of these are alternate spellings given by parents who want an unusual name, but it’s possible some are typing errors by the clerks who recorded the data. There’s no way to determine which are errors and which are intentional, but you should bear these possibilities in mind. Datasets are rarely perfect, and this is especially true the larger they get.
- There are quite a few boys named ‘Jennifer’ in this data. Again, it’s possible some adventurous parents gave boys a traditionally female name, but if you look through names of medium popularity or better, you’ll find a small percentage is always of the other sex. This odd consistency makes it probable that a good proportion of these are also due to errors in the dataset. If you wanted to just consider the girls named ‘Jennifer’, you could filter the Sex column.
The Pivot Tables feature is a powerful tool that allows you to manipulate and explore the data. Here, we’ll use it to find out how many names and births are in the database for each year. First, select columns A through D, so they are highlighted. Then click the Insert Tab’s leftmost button, PivotTable. In the dialogue box that appears, make sure the Table/Range radio button is selected and the accompanying text box reads CA!$A:$D (if you selected columns A through D correctly earlier, this should be the default. If not, type it in exactly as written. The CA is the name of your data worksheet, taken from the CA.TXT filename you started with).
In the bottom of the dialogue box, make sure the New Worksheet radio button is highlighted, then click OK. A new worksheet appears, named Sheet1 – right click on the Sheet Tab and rename it something like ‘Pivot’, since it’s a good habit to always have descriptive sheet names instead of uninformative default ones.
In the pivot field menu on the right, click the checkbox next to the Year field. Year now automatically appears in the ROWS box on the bottom left of that menu, which is exactly what you want. Now click the Births checkbox, and drag the Births that appears in ROWS to the right into VALUES.
Your screen will now look like this:
A few things should be noted here: the title of the rightmost column, Count of Births, is a little unclear. In data analysis, ‘count’ always means the number of rows in a category, regardless of the value in the cells in that row. So what you are seeing here is: for each year in the database, the number of unique male names plus the number of unique female names. You can see that as time progresses from 1910 to 1927, there are more names per year. Does this mean parents are picking more diverse names for their children? Maybe – that’s what you want to find out with further analysis.
Clarity and explicitness are important. Whenever you create a computer document, you should do so with the philosophy that if you open it again six months from now, you will immediately understand what you’re looking at. With that in mind, click on the cell where it says Count of Births and change it to Unique Names.
Bear in mind, when you’re working with pivot tables, the menu on the right will disappear anytime you don’t have a cell of the pivot table to the left selected. If that happens, just select a cell in the table, and you’re good to go.Add a PivotChart
When it comes to quickly understanding data, nothing beats a chart. (Most people call charts “graphs”, but technically a graph is a complicated network visualization that looks nothing like what you’d expect, so Excel properly calls them charts.) Our visual senses are powerful, and are able to immediately understand patterns and trends when they are abstracted into the form of bars and lines.
Make sure your pivot table is selected, then in the Insert Tab, click PivotChart. In the next dialogue box, the default is a bar chart; this will work, but it will be easier on the eyes if you select a line chart, then click OK. You may find it easier if you resize the chart so that the bottom x-axis shows intervals of five and ten years, since we tend to think of years in terms of decades. Your screen should look like this:
Again, we’re seeing an increase in the number of unique male names and unique female names per year. But what if you want to know the number of births themselves? With Excel’s pivot table, that’s easy to do. You could modify your single column, but it is usually more informative to add a new column so you can compare, contrast and calculate.
In the right-hand menu, under Choose fields to add to report, drag the bold checkboxed Births down to the VALUES box in the lower right. You now have two columns, Unique Names and Count of Births (Excel has given this column the same default name it did before). Click the downward-facing black triangle to the right of Count of Births in the VALUES box, and select Value Field Settings from the context menu (the menu that pops up when you right-click). In the resulting dialogue box, change the highlighted Count to Sum.
Your new column’s header name is wrong, so click in its cell and type Number of Births (just “Births” would have been fine, but Excel won’t let you give a pivot chart column the same name as one of the columns it’s based on). A new line has been added to your pivot chart, but because the number of births is so much greater than the number of names, it’s compressed down to near the x-axis. The solution for this is to put it on a secondary y-axis. Click on the compressed series so it’s selected. Right-click and choose Format Data Series from the context menu. Then, choose the Secondary Axis radio button, and click the X in the top right of the Format Data Series panel to dismiss it. Now you should see this:
If you see something different, don’t panic. Go back and follow the steps closely, using this screen as a guide to what you should see.
Let’s study the shapes of the Unique Names line (in blue in the figure above) and the Number of Births line (in orange above). They both have a generally increasing direction, as you would expect, and often move in tandem (especially from 1910 to 1935 and 1975 to 2000). The number of births increases rapidly during the Baby Boom starting around 1940, peaks around 1960, and peaks again around 1990 and 2005.Another Sanity Check
Whenever possible, it’s a good idea to get a second opinion about data: you weren’t involved in its collection or curation, so you can’t vouch for its accuracy. Just because a government department publishes a dataset, doesn’t mean you should trust everything in it 100%. (Please believe me, I speak from experience!)
In this case, it’s easy to double-check. Googling the terms ‘California birth rate’ leads us to the California Department of Public Health, and documents such as this one -- http://www.cdph.ca.gov/data/statistics/Documents/VSC-2005-0201.pdf -- which show the same trends (after 1960, anyway, where the CDPH data starts) as in the Baby Names data. However, it appears that the overall number of births is greater in the CDPH records than in the dataset we’re working on. For example, in 1990, the Baby Names data shows about 550,000 births, while the CDPH shows 611,666.
That’s why it’s a good idea to know your dataset, and read up about how it was collected and what it contains (or what it leaves out). The background information given by the Social Security Administration about this dataset at http://www.ssa.gov/oact/babynames/background.html and http://www.ssa.gov/oact/babynames/limits.html points out that any names with fewer than five births is left out, to protect the privacy of the names’ holders. So it’s plausible that the 60,000 missing births split among people who shared their name with fewer than five other people.
Top Stories Past 30 Days