Six Steps to Troubleshooting Tableau Dashboards

We all know how the story goes… We’re going about our day, building dashboards and fostering insights, when the email comes in. A client asks, “I’m looking at this dashboard and it doesn’t look right, is it accurate?” All of the sudden our day (or even our week) is consumed with troubleshooting to either ensure it is accurate or find the error. Good news! You’ve stumbled upon a handy step-by-step guide on how to check for problems. So let’s begin! Below are the six steps to take to troubleshoot Tableau dashboards

1: Check the data extract/connection

One of the first areas to check is if the extract is updating accurately. Tableau lets you check the last date the extract ran. If the extract failed to update, you can:

1.  See if the workflow (ex. a Tableau Prep workflow) ran properly. If it didn’t, run it and   troubleshoot what caused it to fail.
   
   
2.  Ensure you set up the extract to automatically refresh on the desired schedule.
   
   
3.  If the extract or live data connection is fresh but the data hasn’t pulled into   Tableau, manually refresh the extract and then check to see if there are   network/VPN or server issues causing the data to not connect to Tableau.
   
   
4.  If the data is saved in a location you do not have permission to, reach out to the   admin to give you the access you need.

2: Check the database/data warehouse

If you are connecting directly to a database or data warehouse, you’ll need to check that the data warehouse is updating correctly. There are a couple of ways to do this:

1. Use a SQL querier or Tableau to find the most recently updated timestamp in the table/views. If the data is not up to date, reach out to the database administrator to help identify the problem and refresh the data.
   
   
2. See if you have saved the query to a table or view that doesn’t automatically update. If that’s the case, update the query to automatically refresh when the database does.
   
   
3. Ensure you have the correct access to view the tables in the data warehouse. If you don’t, reach out to the database administrator and request access.

If you go through the steps above and the problem still persists, dive into the schemas/tables to see if the problem is there. See if the schema changed. Did the table get renamed? Did a column get dropped? If any of these changes happened, you can reach out to the database administrator to discuss why the changes occurred. If the changes occurred on purpose then update your SQL or queries to reflect the new tables/columns. If a column name changed, be sure to use the “replace references” function in Tableau to replace the old column with the new one.

3: Check the data in Tableau

The Data Source Pane is a powerful tool that doesn’t always get the love it deserves. If your data looks incorrect but the data sources are up to date, this is the next place to check. 

1. Check your data types. A common error is a field coming in that should be a decimal but Tableau incorrectly assigns it as a whole number. This causes all the values to be whole numbers only, producing incorrect analyses. Another issue is if a field is supposed to be a string but Tableau incorrectly assigns it as numeric. This will cause all string values to come in as nulls. Always check your fields to be sure they are assigned the correct data type.
   
   
2. Check the join. If you are joining two data sources and you have an error on the join, the data type might be the issue as well. In order to join two data types, the join fields must have the same data type (numeric, string, etc.). A common error is IDs; most unique IDs for people (ex. student ID, employee ID, etc.) are numbers but are stored in data systems as strings. Tableau may bring in the ID as a number or as a string, depending on how the underlying data is classified. Change the join fields to be the same type and then the join will work.
   
   
3. Check the distribution. If the report looks incorrect because suddenly an aggregation, such as average or median, changed drastically, check the data distribution. It is common in some fields to assign unknown values as 999. If your data originally assigned unknowns as nulls but then changed to 999s, it can cause your aggregations to spike. Doing a simple histogram in Tableau can help you catch these outliers and handle them accordingly.
   
   

4: See if the report has changed

Many dashboards are updated by multiple analysts depending on the business needs. If a client is asking if a report is accurate, it’s possible the dashboard was updated by other analyst without your knowledge. It’s now your job to try and find what the change was and see if it potentially caused an error:

1. If you have a ticketing system, such as Jira, or another process you use to document changes, check there. If an analyst made a change, it should be in the ticketing system or documentation, and you can either reach out to the analyst or explain the change to the client.
   
   
2. Tableau Server’s built-in version control capabilities allow you to see the previous version, but you’ll have to go through the report manually to see if you can identify any meaningful changes.
   
   
   
   
3. You can see changes in the Tableau XML from version to version if you have access to version control software and know how to use it. You can learn more about Tableau version management in this article by Mighty Canary.

If you don’t have a ticketing system or change-tracking process, consider implementing one. It can be helpful in situations like this to have a system where all dashboard changes are documented so any analyst can see the changes quickly.

5: See if there are “gotchas” in the data

If you’ve checked your data extract/data warehouse and the tables/columns are all accurate, maybe there’s something happening with the data that is coming in. Some common issues are:

1. Nulls. If nulls have started appearing in your data, your calculations may not be accurate. By default, Tableau does not treat nulls as missing but instead ignores them completely. If you have two columns, A and B, and you create a calculated field that is A+B then any row with nulls will not be calculated. Be sure that if you need nulls to be treated as 0s that you wrap your calculations in the ZN function. So your calculation would be ZN(A)+ ZN(B). ZN means “Zero Null” and will force nulls to be 0. See the table below to see the difference in A+B and ZN(A)+ZN(B) once nulls start appearing.
   
   
   
   Nulls will also not appear on axes, instead showing a “there are null values” indicator. Check with your database administrator or data owner to understand why nulls are appearing and be sure that they are handled appropriately, whether that means filtering them out or forcing them to be 0.
   
   
2. Filters. Filters are one of the features that make Tableau extremely powerful but also can be the cause of questions. There a few things that can go wrong with filters:
   1. 1. Order of operations. The order that filters calculate can cause many headaches. For example, if you are using a filter to find the top 10 products by market, you can see in the diagram below that Tableau will filter the top 10 first, then apply any additional filters (such as region). So instead of showing the top 10 products in the south region, it will show which of the top 10 overall products are in the south region.
         
         Filter Order of Operations
         
         
         This also affects Level of Detail (LOD) calculations; Fixed LODs will calculate first, followed by any dimensional filters. But for Include or Exclude LODs, the dimension filter calculates first. So you have to be extra careful with which filters are dimension filters and which are context.
         
         If you think that filters are causing some of the data issues, create a simple cross tab of the data in a new sheet to see what is being filtered and what should be filtered. Adding filters to context can help fix some of the order of operations headaches.
         
         For more information, Tableau has a great article on common issues related to order of operations.
         
         
      2. Extract filters. It can be extremely frustrating when you see that data is missing from your dashboard but you can see the correct numbers in the raw data. You go into the worksheet and check every filter on the filter shelf and still nothing is working. And then you realize – you forgot about the extract filters. The extract filters are on the Data Source pane, not your sheet, and if something is wrong with that filter, your data won’t even come into Tableau. Always check all your filters, both the ones on the sheets and the extract filters on the Data Source Pane.
         
         
      3. “Select all” vs. “Use all.” When using a dimension filter, you can “Select from list” or “Use all.” Since Tableau defaults to “Select from list” most people just stay on this option and then click the “All” button to select everything. Then in a month they notice that the totals on their dashboards aren’t matching the totals in the data warehouse. If you just click “All” on the “Select from list” page, any new values added to the column will not be selected and will be filtered out.
         
         If you want all the values to show in the filter, regardless if new values come in or not, then check “Use all.” This will ensure that all values, even new ones that appear after you created the original dashboard, will be included and not filtered out.
         
         
         
         
      4. Filtering across sheets. Sometimes we need filters to apply to multiple sheets, but if we set this up wrong then a filter from one dashboard might erroneously affect a different dashboard. When this happens, the data is misrepresented because the client cannot see that a filter they don’t expect is affecting the view and they will have the wrong information. Always check that your filters are applying only to the sheets you want/need them to be applied to.
         
         
      5. Hidden data. Hidden data isn’t filtered per se, just hidden from the view. But to the end user, it looks like it’s filtered out and they may think there’s a filter issue. Because hidden data isn’t filtered, it’s a bit harder to see that data has been hidden on a sheet. Always check for hidden data using the Analysis dropdown menu to be sure that isn’t what’s causing the data issue.
      
       
      
      
3. Axes. This has happened to me more times than I care to admit: I open a dashboard, filter the data to what I need, and then all of a sudden my line graph is gone. It looks like there is no data, even though I know there should be. I frantically check all my filters, hidden data, and raw data and it doesn’t make sense – the data should be there. I check my axes because I cannot think of anything else to do it, and that’s when I see it: fixed axis.
   
   If the axis is fixed and new data comes in that is outside the range, it won’t show on the graph. If you need a fixed axis, always make sure the axis is set to the range you need, or set the axis to automatic if you want it to adjust to whatever data is there.
   
   Another axis pitfall comes when you have dual axes. If you are using dual axes that have the same granularity (for example, both are numbers or both are percentages) make sure to synchronize the axes. Otherwise you will have misleading charts.

6: See if the report is confusing

Sometimes the dashboard and data are completely accurate, but there is a misunderstanding of the report on the client’s end. Sometimes it can be a misunderstanding due to dashboard interactivity, and sometimes it can be definitional. One thing to check first is filters; have they selected filters and not realized it? As analysts, we understand the power of filters and use them often. But if our clients are not analysts, filters may not come as easily. They may not even see the filters, or they may not understand how they affect the dashboards. As the dashboard builder it’s important that the reports are set up in a way that it’s obvious if filters are selected. This can involve dynamic titles that change based on the filtered value or adding captions to the sheets with the filter values. This is especially important if your filters are drop downs or hidden in show/hide containers. 

Other issues might be around definitions. Are they understanding the data definitions you are using? As analysts we understand there are a lot of nuances to the numbers we work with. We may know that a certain product should be excluded from the total analysis because it isn’t part of the campaign, or that the official number of employees excludes interns and volunteers, but that doesn’t mean the client knows that. Be sure that the report includes some kind of data dictionary, note, or tooltip for any definitions that could confuse the client and lead to misunderstandings. 

Be Proactive with Mighty Canary