Four Ways to Perform a Top N in Tableau

As a regular contributor on the Tableau Forums, one of the most common questions I see is how to perform a Top N filter. In addition to the basic how-to questions, I also see many people struggling to get their Top N method to meet their particular needs. So, in this blog, I want to share four different methods for performing a Top N filter in Tableau. Along the way, we’ll address some of the pros and cons of each of the methods so that you can make sure you choose the best one for your particular use case.

 

For this blog, I’ll be using a data source that includes 15 customers and their sales. To orient you with the data, here’s a simple bar chart showing the customers ranked by sales.

 

 

Note that Tracy Blumstein and Darrin Van Huff both have the same sales, causing them to be tied at the 10th rank. This is critical as it will come into play later on.

 

The workbook is available on Tableau Public here: Top N Filter Methods

 

Method 1: Top N Filter

To me, the most straightforward method for performing a Top N filter is to use an actual Top N filter. Using our data set, we’ll drag Customer Name to the filter pane, then choose the Top N option. We’ll set it up to show the top customers by Sales. Note: You can specify a number here, but I tend to recommend using a parameter (in my case, the parameter, Top Customers) as that gives more flexibility.

 

 

This is a super easy approach, but it does have some disadvantages. First of all, it doesn’t deal with ties properly. As I noted previously, Tracy Blumstein and Darrin Van Huff both have the same sales, but the top N filter only includes Darrin Van Huff. So, if your data allows for the possibility of ties, then you need to be very careful about using this option.

 

Another potential disadvantage comes into play when you have filters on your view. For example, let’s say we filter our view to only the East region.

 

 

Eleven of the customers in our data set are in the East region so why does this view only show six? The problem lies in the Tableau Order of Operations. As I detailed in my blog on the Order of Operations, this is a pretty common problem. Our Region filter is a dimension filter and, as shown below, Top N filters compute before dimension filters.

 

 

Tableau is first looking at the entire data set and pulling the top 10. It’s then filtering that list of 10 customers down to only the East region. In many (if not most) cases, this is not what you want. Instead, you want to show the top 10 customers within the east region. To address this, we need to change that dimension filter to a context filter so that it computes before the Top N filter.

 



While this example has a simple solution, things can start to get a bit hairy here. Imagine if your measure leveraged a fixed LOD calculation. There might be some situations where you want that LOD calculation to compute before the filter, but want the Top N filter to compute afterward. Unfortunately, you can’t do this because Top N filters compute at the same time as Fixed LODs—you can’t have it both ways. If you run into this situation, then a Top N filter is simply not going to work for you and you’ll need to consider another method.

 

Method 2: Top N Set

Method 2 us a Top N Set. To use this method, create a set on Customer Name then specify a Top N as shown below:

 


Now drag the set to the filter pane and set it to only show “In” values.

 

 

This works almost exactly like the Top N filter. As you can see, it fails to handle ties properly. And, because the set filter computes at the same time as Top N filters, you’ll have the same potential problems with dimension and context filters. One potential advantage of a set is that you can reuse it—either in other views or within calculated fields. In practice, I don’t use this very often, but there could be scenarios where this reuse makes sense.

 

Method 3: Index Calculation

The next method entails using an INDEX table calculation. INDEX essentially returns an ID for each row in a partition. We’ll create a calc like this:

 

Index Filter

// Filter Top N using INDEX

IF INDEX() <= [Top Customers] THEN

    "Show"

ELSE

    "Hide"

END

 

We’ll now drag this to the filter pane and include only “Show” values.

 



Like our previous three previous methods, this doesn’t account for ties. But one advantage of this method is that table calc filters compute near the end of the order of operations so the issues we saw with previous methods will not be a problem here. Of course, there could be situations where you might want something like an LOD to compute after the filter and, in that case, this method won’t really work.

 

The nature of INDEX is that it counts sequentially based on some order. In our example, we computed the table calculation along Customer Name. Because the chart was sorted by sales, the index provided accurate results. But what if it’s sorted alphabetically?

 

 

In this case, INDEX will number the rows alphabetically and produce an incorrect top 10.

 

 

However, because table calculations are flexible, we can change the way it computes to ensure it gets the correct values.

 

 

 

So, while it’s possible to make this work, it just speaks to the inherent complexity of working with table calculations. They are incredibly flexible which is a good thing, but this flexibility comes with complexity. And, if you don’t fully understand what you’re doing, it’s easy to make a mistake.


Method 4: Rank Calculation

Our final method entails using a RANK table calculation. We’ll create a calculated field like this:

 

Index Filter

// Filter Top N using RANK

IF RANK(SUM([Sales])) <= [Top Customers] THEN

    "Show"

ELSE

    "Hide"

END

 

We’ll drag this to the filter pane and include only “Show” values.

 

 

Because the RANK function deals with ties, you’ll notice that we see both Tracy Blumstein and Darrin Van Huff. This is a huge advantage of this method! It is, however, important to note that there are a variety of different RANK functions which work in slightly different ways. RANK_UNIQUE, for instance, assigns each row a unique rank so it will not show ties.

 

Other than how it handles ties, this method is similar to the INDEX option, having many of the same pros and cons. It’s a table calculation so ensuring you compute things properly is essential. However, RANK is a little harder to mess up since it’s ranking the sales measure, not simply counting rows like INDEX. Nonetheless, you still have to be careful to ensure that everything is computed properly.

 

Recommendations

So, which method do you choose? Well, like almost every question in data visualization, it really depends on your situation. I typically start with a Top N filter because, to me, they are the most straightforward and they work for the vast majority of use cases. I only move on to another method if I find the Top N filter to be limiting in some way, such as problems with the order of operations or issues with ties. If ties are a potential concern, then the Rank method is likely the best method as it’s the only one that will deal with ties properly.

 

I hope this blog has been helpful in showing the different methods and the pros and cons of each. Thanks for reading and let me know what you think in the comments section below!!

 

Ken Flerlage, July 11, 2022

4 comments:

  1. Thanks Ken, this is extremely helpful and ties in with a problem I currently have. I've got data for accounts receivable balances by client that I run each week and have constructed a bar chart dashboard that gives me Top N clients and other with 'N' driven by a parameter. I then show the ranking of the clients with the last one as 'Other'. So far, so good. However, what I'd like to do is show is a stacked bar chart (again for Top N and other) bucketed by days outstanding (0-30, 31-60 days etc) The ranking works fine for Top N but introduce 'Other' and it gets goosed up. Any thoughts on why this might happen? Apologies btw in advance for hijacking your thread and also for not posting any screen shots, work are very protective of the data and they also own my licence! Best wishes and thanks again. Craig

    ReplyDelete
    Replies
    1. Probably best to address this via email. Could you email me? flerlagekr@gmail.com

      Delete
    2. I have a similar issue and I am using the ranking method. I want to have the absolute values and next to it, another chart with percentages using another dimension. I can't make the ranks work when I introduce the second chart

      I will try the set method to see if it changes

      Delete
    3. One potential disadvantage of the index method, as I heard in a webinar, is that since Table Calcs are calculated at the end of the order of operations, your viz can get slow if you have lots of elements to rank in your table/chart, since they will live in the canvas before the index filter is applied

      Delete

Powered by Blogger.