In our inaugural Channel Q episode, we wanted to cover questions submitted by our users on Brand Tracking. Due to time, we covered these questions in two episodes, but this post covers responses to all questions. You can watch recordings of the first and second sessions through the links below. 

Watch Episode 1 now

Watch Episode 2 now

The Brand Tracking Q&A questions we received can generally be divided up into 3 categories:

  1. Working with data files and updating between waves
  2. Working with waves and dates
  3. Specific questions around coding in tracking projects.

 

Best practices for tracking data files 

Questions (Episode 1):

  • “How do we set up our Q project correctly from the beginning?”
  • “What are the best practices for setting up and organizing both data and questions in a brand tracking study?”

A well designed and organized data file is important in any Q project but will save a lot of time and prevent future headaches when analyzing brand tracking research. In general, the first wave of a project informs how the subsequent waves should be structured. The fewer differences in later waves, the easier it will be to merge the data.

General data file best practices include:

  • Never change a variable name or code value from wave to wave. For example, if “Sydney” is code 3 in wave 1, it should be code 3 in wave 32. Or if your location question is Q1 in the first wave, then that should stay the same too.
  • If you add new codes, such as adding brands, then use new code values.  Don’t reuse the old ones!
  • Similarly, if you remove a code or question, like removing a brand, then leave the variables or code in your data file, even if it won’t contain any data.

Other best data file setup best practices can be found on the Q Wiki: Data File Setup for Tracking Studies. Best practices for setting up .sav (SPSS) files specifically are also on the Q Wiki: SPSS Data File Specifications.

Updating the data with new waves and fixing errors

Questions (Episode 1):

  • How do we streamline merging data from different waves?
  • How flexible is Q when some of the questions change in a brand tracking study?
  • How do we update the data set when a new wave doesn’t match the same structure as the prior wave?

When working on tracking projects, you should maintain a single master data file that stores all of your data. This means all the different waves and any other, external data need to be merged into this single file. If you’re collecting this data on the same platform, you may be able to get a raw data file directly from the data collection software rather than merging after the fact. 

This blog post walks through the process of merging data files in Q, as well as adding new variables to an existing file. This wiki article has general examples of some tracker-specific errors you may encounter when updating the data. For examples of the errors and how to fix them, please refer to the video of the Q&A session.

Question (Episode 2):

  • How do I import another wave of data, having previously stacked the data using R?

Re-stacking the data in R is the best approach, for consistency, though stacking data using R can create very large data files in some cases. If using the built-in stacking tool in Q is not an option, or the file is causing you hardship, please reach out to [email protected]. We’ll take a look at your file and make any recommendations we can.

Updating banners with new waves

Question (Episode 1):

  • When brands change between runs, do the banners automatically update with the changes?

Depending on the question used in the banner, it may automatically update. Pick One questions or Date questions, for instance, will update automatically. However, a Pick Any question, or any question based on multiple variables, must be modified to account for the differences. More information on creating, editing and updating banners is on our blog and wiki. 

Q also has a great feature where it can automatically compare statistics against the previous period, which is particularly useful for tracking. The Q blog has more information on this feature and how to work with it.

Merging external data into an existing question 

Question (Episode 1):

  • How do you merge external data into an existing question in Q?

Sometimes it’s necessary to add new variables to a tracking study with data created outside of Q. Examples include open ends coded outside of Q or segmentation assignments. In this scenario, you should merge in the new variable into your master data file by matching up to the IDs in your file. However, this newly added variable will only have the new data as a separate variable. If it needs to be merged into another question, you will create a third variable that brings together the data from the two existing variables.

For more information on merging data files, please see this Q Wiki article: Merging Data Files.

Updating deliverables and reports

Questions (Episode 1):

  • How to automatically update charts and graphs in PowerPoint on an ongoing basis?
  • Does Q account for small changes when exporting data into PowerPoint, or will any changes made to the original setup need to be inputted/amended in PowerPoint manually?
  • What is the best way to use Q to help make brand tracking presentations faster to update?
  • How do I set up reports that will be used with each new wave of data?

Updating an existing PowerPoint or Excel deliverable with a new wave of data simple in Q. With the deliverable open, Export to PowerPoint or Excel from Q and select Update when prompted. Q takes care of the rest. More information about automatic updating of PowerPoint or Excel is in this blog post.

In addition to automatically updating tables and charts, you can also create text in Q that automatically updates from Q. You can read more about how to use this in your Q project on our blog. 

Question (Episode 1):

  • When using Q to automatically update a large number of charts in a presentation, what can you do to stop the app from crashing/ having an error and making the powerpoint unstable?

Sometimes, the export is interrupted – e.g. by accidentally clicking into the outputs as they generate. There are also processes that Q cannot control that can cause the export to go awry. The best thing is to click Export and leave the computer until it’s done.

The whole exporting system is currently under review, and we’ll be introducing a new way to do this that will not need either PPT or Excel to open at all when exporting, while still maintaining the links between Q and Office.

 Question (Episode 1):

  • How do you find out which powerpoint chart is linked to which Q table?

Unfortunately, the code that links the chart in PowerPoint with the table in Q is hidden in the programming in PowerPoint and isn’t visible in either PowerPoint or Q. 

Questions (Episode 1):

  • Some large banners become slower to calculate and load. Are there any tricks to making this more time-efficient?
  • What are the best ways to account for large data sets?

While there are no specific tricks to this, the easiest solution is to have smaller banners with fewer columns. Very wide tables are rarely read or useful to end-users. Instead, focus on what’s important and consider editing your tables.

Some tables can take a long time to compute, so it’s also a factor of the number of cases and whether you’re using complicated Rules on the table. The Q wiki has several recommendations on working with large files and how they can be structured more efficiently.

Questions (Episode 1):

  • How to best set up your tables to quickly identify changes from one tracking period to the next?
  • If there is a difference in performance for a new wave of data, how to quickly identify what groups those differences are coming from without having to go through a lot of trial and error?

Though no two tracking projects are alike, Q does have some tools to sift through many crosstabs to help you find changes and insights. This blog post goes into greater detail about the tools. One of those tools is Smart Tables, which can be used to quickly generate sets of tables that compare significant results within two periods. 

Working with Waves and Dates

As your tracker will be updated over time, you may want to create filters that identify certain time periods, wave specific weights, or other calculated variables. Q can automatically recalculate all of these when you update your data.

Question (Episode 1): 

  • How do I create time filters that automatically update when I include a new wave of data?

On the Variables and Questions tab, you can right-click and select Insert Variable(s) >  Binary – Complicated Filter then setup logic for the time period you’d like. More detail can be found in our blog post on How to Build Tables that Automatically Show Results from the Latest Period.

Question (Episode 2): 

  • If you’re using rolling averages (not that I’ve done that since the 90s),  how to easily set those up in Q?

There are a couple of ways to do this outlined in this blog post, but the easiest way for a tracking study is to follow the second method. It uses a Date question in your crosstab and Time Series Analysis function. Alternative methods involve creating new variables each wave of the project, and although more flexible, will take a bit more effort.

Question (Episode 2): 

  • …how to use 2 different weighting [sets] when comparing waves?

It depends on whether you have the SAME targets for each wave, or if you have DIFFERENT targets for each wave. For general information on creating weights in Q, you can see our wiki page here or download our ebook. 

If you’re using the same targets for each wave, you will set up your targets as normal and select your Date/Wave variable from the Recompute for each dropdown.

If you’re using different targets for each wave the process is a bit more involved depending on if you’re weighting on more than 1 question or if your waves are weighted equally or not. If you want to keep your waves unweighted, but weight questions differently across them, you will create a separate weight variable for each wave. You will use your Date/Wave variable in the Target column question(s) dropdown and put in 0s for the targets for the waves not associated with the particular wave’s weighting you are creating. Then you will create a JavaScript variable to combine each wave’s weight into a new weight variable to use with your table. An example of how to combine weight variables using JavaScript is on our blog here. 

Coding with Tracking Studies

Coding is one of the more tedious aspects of any type of study and is multiplied when you have a tracking study. Luckily you can reuse previously coded answers and other automated coding tools in Q to code your new responses with less effort.

Questions (Episode 2): 

  • How can we analyze and code open ends?
  • We are looking at including verbatim questions in our brand tracking, how do we analyze the verbatim in Q?
  • How best to clean & code open ends for top of mind awareness questions?

For analyzing verbatims, typically a researcher will code the text into a series of codes (or categories), then use those in their analysis. Q has a built-in coding tool where you can manually code your verbatims in a more streamlined fashion that you are used to coding in Excel. More detail on this tool and some guides on how to use it are listed on our wiki here. When you update your tracker with a new wave of data, the new responses will be coded using your previously coded responses as the rules (known as a code frame). Then all you will need to do is code the open-ends that weren’t in previous waves. You can do this by right-clicking on the coded version of your question on the Variables and Questions tab and selecting Edit Code Frame.

Questions (Episode 2): 

  • How can we quickly or automatically code open-ended responses?
  • How to automatically code open enders into the same code frames already used?

Q’s automatic categorization tools work nicely with trackers where you want a more automated approach to coding new responses or even your first wave of data. 

  1. Create > Text Analysis > Automatic Categorization > Unstructured Text uses machine learning to analyze your text in bulk and code it for you. If you already have some of your text coded manually using our coding tool (either from a previous wave or a subset of the data), you can feed this into the analysis using the Existing Categorization field to help “teach” the algorithms how to automatically code any new or uncoded responses.
  2. Create > Text Analysis > Automatic Categorization > List of Items is designed for brand lists you may want to use for your top of mind questions but works for any type of list. In the output created you can click on Diagnostics and view the Variant Suggestions table to see how each text was categorized. Copy this table into Inputs > REQUIRED CATEGORIES > Add required phrases and variants to confirm the list of codes and their variants for the analysis.
  3. Create > Text Analysis > Automatic Categorization > Entity Extraction identifies people, places, and other entities automatically from data.

To save your automated results, click on the output and select Create > Text Analysis > Advanced > Save Variables > Categories.

Question (Episode 2):

  • How to fix open-ended coded multi-select questions so they don’t fall apart and can be merged with previous waves more easily

Although I’m not sure what’s happened in this specific case, please reach out to [email protected]. You should review the coding to make sure there’s nothing new that needs to be assigned. Right-click the coded question from before > Edit Code Frame. Then you can code the remaining responses.

Other Questions

Below are some other questions that were submitted by our users. 

Question (Episode 2): 

  • In an existing question with several nets, how do you handle adding new choices to the question in the next wave without having to redo the NETs?

The NETs an existing question will not know that you’ve added new choices, such as brands, or how to group them, regardless of the question type. Consequently, they need to be updated when you import the data that has a new choice in it. You can set up variables that do this programmatically, but if those variables are not numeric, they would need to be updated on a wave-by-wave basis.

The exception to this is numeric questions, such as age, which can be banded using scripting and will automatically incorporate the new data. More information about banding numeric variables is on our wiki here.

Question (Episode 2): 

  • How can I build custom tables that pull in different metrics that belong to different variables?

If you have two or more tables that you want to combine and the questions are different types, you can use Automate > Browse Online Library > Tables > Merge Two or More Tables to combine them. If you need to combine bits and pieces from multiple tables, you will need to add an R Output to your report and create R code. We have a nice tutorial on how to do this on our blog here. Please keep in mind, both of these options will not bring over things like significance testing, so use with care. If the variables are the same type, such as a set of numeric variables, you can combine them using Set Question.

Question (Not included in Episodes 1 or 2): 

  • … using R to generate tables?

This is a very broad question, and so the best place to start is to review the documentation on our wiki Using R in Q. For specific resources on how to generate and manipulate tables in Q, please see our list of blog posts here.

Question (Not included in Episodes 1 or 2): 

  • What are the latest developments in charting automation? 

As demonstrated in the webinar, you are able to export and update charts in bulk from Q to Office by simply highlighting multiple outputs or folders in your Report tree. As for visualizations that we offer, we are continually updating our visualization library to meet our users needs. You can check out the most up-to-date list of visualization automation under Automate > Browse Online Library > Visualization.

Question (Episode 2):

  • How can Q be used for Factor and Key Drivers Analysis – including presenting the results?

There is some excellent documentation on the Q-wiki in the form of a webinar on this topic, which you can find here. There is also an accompanying eBook on this topic, which is free to download from the Q Website, here.

Question (Episode 1): 

  • Is there any way to make latent class analysis or a cluster analysis in a new wave without affecting the values of the previous wave and follow the same logic?

The short answer here is “Yes”. With a Latent Class Analysis, as long as the variables in the new data are exactly the same and there are no unexpected differences, then you simply update the data file with new data. If there are differences, then other methods are best, and it may be necessary to use predictive analytics (e.g. Linear Discriminant Analysis) to work out an approximation of the segments from the existing data. Our blogs have some examples of how to apply a typing tool in Q and how to export an LDA function to excel.