2021: Week 36 - Solution

 

Solution by Tom Prowse and you can download the workflow here. 

There have been some big changes to our lives in the past couple of years, so this week we looked back at what people have been searching for on Google and how this has changed. 

Step 1 - Average Index

First we are going to focus on the Timeline data table, where we want to calculate the overall average index for each search term. Before we can calculate the average, we need to transform the table by pivoting the main search terms from columns to rows. We can use a Wildcard pivot for this by using the matching pattern 'World': 

After the pivot, we have the search terms in a single column but we need to do some tidying to remove the 'Worldwide' part. This can be easily done with an automatic split, or by using a custom split to return the first occurrence before the ':' separator. 

Once we have tidied and renamed some of the fields, our table should look like this:

Now we have the table in this format, we can calculate the overall average index by using a Fixed LOD. We use an LOD here instead on an aggregation tool because we want to keep all of the rows of data.

Avg Index 

Then we can round the averages to 1 decimal place so that it is more manageable: 

Avg Index 

ROUND([Avg Index],1)

This is step 1 complete as we have calculated the overall average index for each search term:

Step 2 - Earliest Peak

Next we want to find out what the earliest was for each of the search topics to hit their peak. We are going to assume the 'peak' is when the index is the highest, therefore we first want to identify each of the max index values, and then find the earliest week that this occurred. 

1. Max index

The first calculation we need is to find the maximum index for each search term. We can use the following LOD to find this:

Index Peak 

Now that we have each of the 'peak' values on each row, we can use a filter calculation ([Index Peak]=[Index]) to identify the weeks where the index was the highest. 

This leaves us with a single week for Online Streamer and Pet Adoption, however there are various weeks where Staycation hit it's 'peak'. Therefore, we need to use a similar technique to identify the first (min) week and then filter to return only that week. 

First Peak 

Then we can use the filter to remove any weeks that aren't the first one ([First peak]=[Week]).

After these changes our table now has a single row for each of the search terms, and the first week that they hit their highest index: 

Step 3 - Yearly Average Index

We now want to calculate the average index for each year, however the year starts in September therefore we need to do some changes before converting the years. On a new branch, from the step before we calculated the Avg Index, we need to classify each of the dates using the following IF statement: 

Year 

IF [Week]<date("2017-09-01")

THEN '2016/17'

ELSEIF [Week]<date("2018-09-01")

THEN '2017/18'

ELSEIF [Week]<date("2019-09-01")

THEN '2018/19'

ELSEIF [Week]<date("2020-09-01")

THEN '2019/20'

ELSE '2020/21'

END

Then using this year we can calculate the yearly average by using an aggregation tool: 

And then we can round the averages to 1 decimal place:

Yearly Avg Index 

ROUND([Yearly Avg Index],1)

As we are only focused on comparing this year to last year, we can filter out the remaining years so that only 2019/20 and 2020/21 remain. Our table now looks like this: 

Step 4 - Lockdown Fad or Still Trendy

We can now compare whether the trend has increased or decreased since last year by comparing the avg index this year compared to last year. 

The first step is to bring each of the years averages into the same row by using a Rows to Columns pivot: 

Now they are on the same row, we can use the following calculation to classify the lockdown fads or still trendy categories: 

Status 

IF [2020/21]>[2019/20]

THEN "Still trendy"

ELSE "Lockdown Fad"

END

Our table should now look like this: 

Step 5 - Country Breakdown

It's now time to focus on the country breakdown to see if these trends are similar across various countries. By using the Country Breakdown table as a new input, we can start by cleaning the table a little bit by removing any null values from the Pet adoption: (01/09/2016 - 01/09/2021) field. 

We then want to pivot each of the search terms so that they are in a single column, by using a columns to rows pivot: 

Then we can tidy the fields by using an automatic split and then renaming the field names so that we have the following table: 

Finally we can filter the search terms so that we only have the countries with the highest percentage. This uses a similar technique to earlier where we identified the highest percentages using an LOD and then use a filter to leave only the highest.

Highest % 

Then use this filter to retain only the highest % - [Highest %]=[Percent]. 

Our countries should look like this: 

Step 6 - Bringing Everything Together

The final step this week is to bring each of the three branches together into a single data source. First we want to join the Trend Peak branch with the Yearly Avg branch by joining on the Search Term field: 

Then after this join we can join the country breakdown branch, again using the Search Term field: 

After these joins we have our final output that looks like this: 

After you finish the challenge make sure to fill in the participation tracker, then share your solution on Twitter using #PreppinData and tagging @Datajedininja, @JennyMartinDS14 & @TomProwse1

You can also post your solution on the Tableau Forum where we have a Preppin' Data community page. Post your solutions and ask questions if you need any help!