Author Archives: Tabitha

Hidden motivations

MotivationsWhen you know your audience’s motivations, you can create a site that speaks to what the audience wants to achieve. If you don’t know their motivations, you may be able to build a perfectly fine site, but it will lack persuasive power.

Motivations are hard to pin down, though, because they are fickle and vary from person to person. And while the goals and challenges of users are often discussed, motivations are sometimes overlooked.

Definitions

A goal is what the user wants to accomplish. A challenge is what might stop them from accomplishing their goal. And a motivation is why they want to accomplish the goal.

All straight forward. Except the motivations. Those can be very complicated.

Motivations

Motivations are layered. There are motivations users are happy to talk about. These line up well with the concept of benefits. On a deeper level, though, there are motivations users may be less comfortable discussing and that they may not even be fully aware of. However, these are an important part of making the buying decision.

Suppose Tom buys a designer laptop bag. If you asked Tom why he bought the bag, he might say something like ‘I need a bag to carry my laptop around in and these bags look great’.

Tom’s goal is to buy a bag. The challenges he faces are to decide which bag, determine whether his laptop will fit in it, and how to make the purchase. His motivation seems to be practical (carrying around his laptop) and aesthetic (the bags look great). This goal, challenge and motivation are all easy to design for: Create a simple site with nice pictures, specifications and a clear purchase flow.

But Tom has another unstated motivation. The smartest, hippest people Tom works with all have this designer laptop bag, and he is using the bag as an indicator that he fits in with that social circle.

So the initial assessment of motivation doesn’t run deep enough. Tom will happily tell you he is buying the bag for practical and aesthetic reasons. Tom is comfortable discussing this level of motivation. He has another, deeper motivation, though, tied to social status and connection. He may not even be fully aware of the deeper motivation, and would be less likely to discuss his personal situation.

But this deeper motivation of Tom’s is important. If it holds true across a large enough audience segment, then it should influence the design and branding of the site. For instance, it might move the site from being practical to being more status aware. The exact look and feel would need to be adjusted for the audience.

It is worth noting that some of Tom’s motivations are not based on actual qualities of the bag, but upon his environment. The company that sells the bag may have contributed to the environment through marketing, but it has particular impact on Tom because of his social situation.

Conclusion

Motivations are hard, especially the ones that people are less likely to talk about and the company didn’t intentionally create. You can sometimes get at them indirectly, e.g. through clever survey questions or by looking at how people behave rather than what they say. Digging down into motivations, though, typically requires a leap of intuition and then checking to see if the data backs you up.

Identifying motivations, especially ones below the surface, is important for building a successful site. If you know audience motivations, you can be much smarter about how you present information and build a more persuasive site.

Why you need a hypothesis

why you need a hypothesisWhen thinking about site tests, there are two approaches you could take.

The first is to just chose something to test. This is appealing; it would fun and easy. Whatever caught your eye, you could change and run as an A/B test against the original version. Sometimes you would be lucky. More often, though, you would be wrong. The changes would be coming from you rather than from what the audience wants. A site change’s success is entirely dependent upon the audience’s reaction. If the audience isn’t happy, your site change will not work.

The second approach is to think about the details of the test and form a hypothesis. When you set up a hypothesis you consider your audience’s point of view.

A hypothesis can be thought of as a testable short story that describes user behavior when a change is introduced. For instance, a hypothesis might be, “Changing the hero image copy to focus on the design benefit rather than technical benefits will increase revenue by 10%”. It states what’s changing (the copy) and the predicted audience behavior (10% increase in revenue).

While it doesn’t appear in the hypothesis statement, the most important part of constructing a hypothesis is why you think it will work. If you can’t convincingly answer that, then your hypothesis is very unlikely to be successful.

You can answer why a hypothesis will work because you are knowledgeable about your audience. Testing with a hypothesis is a great way to increase that knowledge. If you test new copy and it works, it enriches your existing ideas about your audience. If it doesn’t work, it adds to your knowledge of your audience. You had an idea as to why it should work but that idea was wrong. So now you need to reconsider what you know about your audience. You can use this information to make your future tests stronger.

Without a hypothesis, a winning or failing test lacks context. It’s hard to interpret because the test wasn’t created with the audience and site in mind. Instead of enriching already known information, the test results are isolated data points.

Hypotheses are also useful because you can use them to prioritize site changes. If you have many changes you want to make to a site, look at what hypotheses are most worthwhile and run those tests first. You get the most impact for your testing efforts.

Finally creating hypotheses is a good way to get support from the people you work with. A series of hypotheses that ties into your site strategy will give people confidence that you know what you’re testing, why you’re testing it and what outcome you anticipate.

Choosing between hypotheses with confidence

A common site optimization problem is that you have too many hypotheses and not enough time to test them all. So the hypotheses have to be prioritized. People prioritize using scope, predicted impact, timeline, dependencies and risk. An often overlooked factor, though, is confidence.

Confidence is simply a measure of how certain you are that the test will succeed.

A hypothesis that is well supported by evidence is more likely to be successful, and so you’ll have a higher confidence level. The more information you have to go on, the better you can refine your hypothesis and anticipate errors. The variables that could upset your results are decreased and the hypothesis is more precise. Knowing more doesn’t guarantee that your hypothesis will be successful, but it does make it more likely.

A hypothesis with less data to support it is less likely to be successful, and so you’ll have lower confidence. Your hypothesis is less trustworthy because the information that you are unaware of can radically change the interpretation of facts. Knowing less doesn’t mean that your hypothesis will fail, but it does make it more likely.

As a general rule, more information means greater confidence in a hypothesis and less information means a lower level of confidence. Depending on risk tolerance – both personally and within the company – low or high confidence in a hypothesis can be a significant prioritization factor.

So what do you do if you have a fantastic idea, but not a lot of data to back it up, i.e. your confidence level is low? My favorite solution is to run thought experiments to find evidence to prove, refine or disprove the hypothesis.

Sometimes, though, even if you can’t find a lot of information to back up a test prior to launch, it’s still worth running. It comes down to what the rest of the testing queue looks like and what a successful or losing test would tell you.

Elements of a hypothesis

Hypotheses are a necessary tool for site testing. A hypothesis can be thought of as a short story that predicts how users will react when a site change is introduced.

Hypotheses are valuable because they keep you focused. If you can’t say what impact you expect a change will have or why you believe the change will work, it’s a good time to dig into the evidence and make sure you really understand what’s going on. Writing a hypothesis is an early test of your idea to ensure it holds together before pushing changes live on the site. Hypotheses also make it easier to prioritize tests and share ideas with other people in the company.

The elements of a hypothesis include: the area or functionality; what you want to change; why it matters to your audience; the metric you’re tracking; and the predicted outcome.

Let’s run through a quick example of how this all comes together.

Suppose you run a social media platform. You want your audience to interact with the platform more frequently both as creators and viewers of content. You believe the best way to accomplish this is to increase the number of photos people upload.

Area or functionality: Your audience’s interaction with the site’s interface, specifically uploading images.

What you want to change: The strength of the button for image uploads so people are more likely to see it.

Why it matters to the audience: Focus groups have indicated that customers love sharing photos over the social media network as it is quicker and easier than writing an update but it still feels like keeping in touch. However usability tests have shown that customers have trouble finding the button with which to upload photos. (Why you think it matters won’t be in the final hypothesis statement but it’s a very important step when building hypotheses. If you can’t convincingly explain why you think an audience will react well to the change you are proposing, your hypothesis is more likely to fail.)

Metric: The number of photo uploads divided by the number of total updates on the social media platform. Right now, photo uploads account for 3% of all updates.

Predicted outcome: An increase in photo uploads so they account for 5% of all updates.

Pulling all these pieces together, the hypothesis is: “Making the photo upload button stronger will increase the share of photo uploads to 5% of all updates.”

If you have a great idea but are having trouble creating the elements of a hypothesis, it may be because you need to narrow down what you’re testing. It is hard to create an effective hypothesis, and effective site tests, around an idea that is too broad. Another possibility is that your idea is sound but you need to refine it a bit more before you can act on it. By digging into available evidence, you can develop more insight into what will work and why.

Google Analytics: Tracking logins

Software as a Service (SaaS) companies often place a link to the login in the top navigation of the marketing site. Both current customers and visitors to the marketing site come to the site and the two populations get mixed together in the Google Analytics reports. This can make it very difficult to analyze your data.

In this post, I’ll talk about why being able to separate these two audience groups matters and how to separate them in Google Analytics.

Why it matters

When current customers are mixed with visitors to the marketing site, it can cause problems when you analyze the data. For instance:

  • An important metric on marketing sites is the eCommerce conversion rate, i.e. the number of conversions divided by the number of sessions. It’s a useful metric when taken in context with other site data as it points to the effectiveness of a site. If a lot of people are coming to the marketing site to login, then it artificially lowers the eCommerce conversion rate. This is unfortunate as it makes the metric unusable.
  • On the flip side, if you don’t have a way to isolate the sessions that include a login, then it can be difficult to analyze the Google Analytics reports. Even a simple question like ‘How many sessions to the site included a login?’ can be very hard to answer.
  • In some cases it can also be useful to segment out how the users who are currently customers and logged in interact with the marketing site – this is especially true if the product includes upsells that direct customers to the marketing site.

By segmenting out the users who login, you get a better handle on how different audiences interact with your site.

Custom dimensions

Custom dimensions add non-standard data to your Google Analytics reports (see the earlier post on determining the data set for background on how Google Analytics is structured).

A custom dimension is set at the property level. To set up a custom dimension, go to the admin section, in the property column click on Custom Definitions, and then select Custom Dimensions.

Unless you’re a premium user, you can only add 20 custom dimensions. When you set up the custom dimension it will ask you for name, scope and whether it should be active or not. How you set up the custom dimension will depend on your specific needs.

For instance, setting the scope to user sets the custom dimension at the user level. Even if the user does not log in on future visits, they will still be grouped into the custom dimension. This makes sense if you’re looking at binary situation where someone is either a customer or not a customer.

On the other hand, if you set the scope to session, then the custom dimension only lasts as long as the session does. If the user ends the session and then comes back, the second session will not be captured in the custom dimension. This might make sense if your users move fluidly between customer and prospect.

In order to track custom dimension, the developer will need to make changes to the code to capture the action you are interested in.

You can construct filters using custom dimensions. This is extremely powerful. You could set up a view that includes only logins by filtering on a custom dimension that tracks logins. You could use the same idea, but with an exclude filter, to remove all the logins and just view the visitors to the marketing site.

Using custom dimension to track logins is an elegant solution, and it’s embeds the data in the reports. For more details on how to set up custom dimensions, see this Google Analytics article.

Segments

In some cases, you won’t be able to use custom dimensions to track logins. There are only 20 custom dimensions available (unless you’re a premium customer) and perhaps they’ve been used for something else. Or perhaps you can’t get the resources to make the code changes that are necessary. Or maybe you can move forward with a custom dimension but you need to look at historical data. If you find yourself in a situation where you can’t employ custom dimensions, you may need to work with data at the report level using segments.

The exact set up of the segments will vary depending on the site, but the idea is to use the URL structure to segment by user. (If you’re not familiar with how to build a segment, check out this Google Analytics help doc.)

For instance, suppose your marketing site is on these three pages:

  • www.example.com/features
  • www.example.com/pricing
  • www.example.com/signup

The pages for customers behind the sign in are:

  • www.example.com/dashboard
  • www.example.com/app
  • www.example.com/stats

In this case you can set up a segment that filters using the dimension Page so you only include users who have visited /dashboard or /app or /stats, i.e. users who must be logged in because these pages can only be visited when someone is logged in. An exclude filter on the same lines would remove people who have logged in.

There are some downsides to this method, however.

Adding segments manipulates data that has already been passed into the report. Over long time periods or with large data sets, Google Analytics often uses sampled data when a segment is applied. Sampled data can, in some cases, be quite inaccurate.

These segments are also fairly difficult to get right. For instance, you have to capture every path in case someone has bookmarked a link directly into the app. Likewise, the URL structures that a user accesses when they log in has to be unique from the URL structure that exists for non-logged in users. For instance, if the marketing site and the application share a path in places, you may not be able to distinguish between the two audience groups. You also have to update the segment regularly if the URLs change, and this can be difficult to stay on top of.

Upshot

Custom dimensions are the better choice for separating logins from other audience groups visiting your marketing site. However, in some cases you will have to work with the data once it’s already been passed into the reports. When that is the case, segments can be a very powerful tool.

Thought experiments: Groundwork for successful site changes

By running thought experiments before making a site change, you increase the chances that when you roll out the change it will be successful.

When you run a thought experiment, you attempt to prove, disprove or refine an idea with existing evidence. You can iterate on ideas very quickly, thought experiments are extremely cheap to run and, since they are not audience facing, they are also very low risk.

The evidence you dig into can be from a variety of sources, e.g. site analytics, focus groups, surveys, etc. The goal is to understand why a change would or would not work.

To walk through an example… Suppose the company you work for sells artwork that ranges from cheap prints to fairly expensive reproductions. Most of the orders are for cheap prints, so the company focuses on these low value, high volume sales. You have been instructed to increase revenue by promoting the cheap prints more effectively.

As a thought experiment, you decide to test whether the low value sales are what the site should primarily promote.

When you look at the order numbers, you find that most purchases are in fact low value. But when you look at the revenue, there are two peaks. One peak corresponds to the low value sales. The other peak corresponds to the high value sales. While there are far fewer high value sales, the margin is so much higher they bring in a lot of revenue.

This suggests that instead of one primary audience behavior there are two – one behavior pattern is to purchase low value products and the other is to purchase high value reproductions.

To further refine your thought experiment, you pull out information on who these high value purchasers are. You discover that they are concentrated in urban areas, specifically Los Angeles, San Francisco and New York. The types of artwork that they most frequently purchase are large, abstract paintings. The paintings are purchased by individuals, not businesses. It appears likely that people are purchasing these paintings to decorate their apartments or homes.

By following a thought experiment you’ve found out a lot about a profitable audience segment that hadn’t been identified. You also have some new ideas for how to increase revenue, e.g. make high value artwork more visible on the site, expand the types of artwork offered, start a weekly feature about how to decorate your urban apartment with artwork.

Thought experiments can help you find patterns and audience behaviors that have been overlooked. By running thought experiments before making site changes, your site changes are more likely to be successful because they will be informed by evidence rather than assumptions.

How to decide what to test

Imagine this: You have just finished leading the installation of a new testing system on the company’s marketing site. You’ve learned the new testing interface. You’ve done your research on A/B and multivariate testing and which to use when. All eyes are on you to improve revenue generated by the site.

Now you need to decide: What to test?

What are you testing for?

Before you start developing any tests, decide what you are testing for. Make sure you know the company’s goals. What is it that you are solving for by testing? What metric needs to improve for a test to be judged successful? Why does this metric matter? People get into trouble by not thinking this through before testing.

For instance, perhaps the VP directs you to increase the average time on page. You run some tests and successfully increase average time on page by 20%. Your site changes are judged a failure because even though you did what was asked – increase the time on page – the number of leads didn’t increase. The VP made an assumption that the longer a customer was on the page, the more likely he would be to submit his information. This assumption was wrong. What you should have been testing for was how to increase the number of leads.

To avoid this type of situation it is very important to understand what metric you are testing for and why.

Once you have your metric sorted out, the next step is to choose what to test. There are a lot of facets to deciding what to test, but to get started I’ll run through a couple of basic approaches: Best practice tests and audience motivation tests.

Best practice tests

Best practice tests involves fixing obvious usability issues. The goal is to get out of the way of your audience. It’s not the most exciting form of testing, but usually there are lots of areas to improve, especially if the site has never been tested before.

There are many ways you can find usability issues on the site. For instance, you can talk to the lead designer and see what she would like to improve. You can go through the site and ask yourself, ‘If I’d never seen the site before, what would confuse me?’ You can run usability testing and see where people hit snags.

This type of testing tends to be straightforward and generate good results.

Audience motivation tests

Audience motivation tests are more interesting, but typically harder to get right. While best practice tests are about getting out of your audience’s way, audience motivation tests are about lining the site up with what your audience wants to do.

To start thinking about audience motivation tests, consider these questions:

  • What area are you testing? What is the metric?
  • What does the company want the audience to do and why?
  • What does the audience want to do and why?
  • What do you want to the audience to do next?

The answers to these questions vary widely, but as an example:

  • What area are you testing? What is the metric? The product pages. Revenue.
  • What does the company want the audience to do and why? The company wants the audience to purchase their handmade artisan clothing to generate revenue and so forward the company’s mission statement of supporting local craftsmen.
  • What does the audience want to do and why? The audience wants to purchase the clothing because it is attractive and distinctive, which feeds into their sense of self, and because they are emotionally committed to projects that help small businesses. These purchases also support how they want the world to be.
  • What do you want to the audience to do next? To add a product to the cart.

If you have compelling answers for all of these questions your site changes are more likely to be successful. For instance, in this case perhaps you need to have better pictures of the product, because the audience is motivated by the clothing being distinctive and attractive. Or perhaps you need to include information about the person who made the clothing because this will reinforce the customer’s ideas about helping small businesses and about how they want the world to work.

Audience motivation testing is generally more difficult. It requires research and imagination. But, in the long term, you will build an understanding of your audience and their motivations, which will help you construct a more effective site. After all, without an audience, your site is just a collection of code.

Google Analytics: Working with the data set

In my previous post, I wrote about how the data set is determined in Google Analytics at the view level and by how the tracking code is implemented. In this post, I’ll look at some my favorite tools for working with data in Google Analytics.

Working directly with the data occurs at the report level. At this level, you can change how the data is displayed and filtered, but you can’t alter what data is available.

There are many ways to interact with data in the reports. While I’ll run through some of the tools I find most useful, this is not a complete list. I’ll dig into other tools in later posts.

Working with data

Regular expressions

Learning how to use regular expressions is necessary if you want to use Google Analytics reports to their full potential. With regular expressions you can match and filter data.

For instance, if you are looking at the Geo location report and want to look at just the sessions from California and New York (and exclude the sessions from the rest of the states) you can filter down to just these two states with the regular expression: California|(New York)

Alternatively, if you want to look at just the stats for the pages in the top level navigation, you could write a regular expression to filter on the Site Content -> All Pages report so only those pages are returned in the results.

With regular expressions you have a concise, exact way to extract the data from the reports that you’re interested in. Regular expressions can be used throughout the Google Analytics interface.

Segments

Segments are one of the most useful tools for working with data in the Google Analytics reports.

With a segment you can isolate and examine data. Segments can be  applied across most of the reports Google Analytics offers. For instance you could compare two segments – direct traffic and paid traffic – in the All Pages report to find out if there was a difference in what pages the two audience groups are interested in.

Access the segments through the interface at the top of the page:

Segment Chooser on Reports

When you click on “All Users” or “+Add Segment” a menu comes up that lists all your segments.

Google Analytics includes some segments out of the box, such as Bounced Sessions, Mobile Traffic and Returning Users.

If the segment you want is not already available, you can build your own custom segment. For instance, you can build a segment that excludes all referrals or includes users based on when they  first visited the site.

Segments are incredibly powerful and useful, but they do have some drawbacks, the most significant is that the data is sampled over longer time periods. If you’re looking at a report with relatively low numbers, this can skew data badly over relatively short periods of time. Even for larger data sets, it can still be a problem. Fortunately Google Analytics lets you know when the data is being sampled so you don’t get caught unaware.

On-page filters

You can narrow down the results set by using the filter at the top of many reports:

Google Analytics on-page filter

You can enter either text or a regular expression into the field. If you want more control, to the right of the box, there is an advanced link. This opens up:

On page filter expanded

The advanced tool gives you more options, including being able to exclude as well as include data.

The filters are a fairly simple tool but immensely helpful for analyzing data as you can look at just the information you are interested in. You can do a quick and dirty analysis without ever having to open a spreadsheet.

Custom reports

Custom reports are a good way to extend Google Analytics capabilities to create exactly the reports you need.

To access custom reports, click on Customization in the upper navigation. When creating your report, you’ll need to add dimensions and metrics.

The difference between dimensions and metrics is often confusing for new users. A dimension is what you want to track (e.g. a page, new users, etc) and a metric is the measurement associated with what you want to track (e.g. the number of pageviews, the number of sessions, etc). This can be a bit confusing at first but it becomes self-evident with practice.

Once you have chosen the dimensions and metrics, you have the option to add a filter. For instance, you could filter on a specific medium such as organic, so only sessions with a medium of organic will appear in your custom report.

Finally you choose which view(s) to associate the custom report with.

Custom reports are a bit tricky so you always have to be sure you understand what you’re asking Google Analytics to do. For instance, if you have a dimension of page and a metric of users, the total number of users reported by the custom report will be correct. But if you add up the number of users associated with each page it will be significantly higher than your total number of users. This is because a user can visit more than one page. What Google Analytics is reporting is correct, but it’s easy to fall afoul of it if you don’t understand the output.

Conclusion

Over the last two posts I’ve looked at how the data set is determined in Google Analytics and reviewed some of the more useful ways the data can be worked with. Understanding when you’re changing what data is being pulled in and when you’re simply working with what’s already available is important as you will be less likely to accidentally damage your data set and you will find it easier to get accurate answers with Google Analytics.

Google Analytics: Determining the data set

In Google Analytics you make choices that change what data gets passed into your reports. Understanding where and how those choices get made is important because they have a profound effect upon the accuracy of your reports.

In this post, I’ll focus on changes that determine the data set that gets passed into Google Analytics’ reports. In the next post I’ll take a look at how you can work with the data in the reports.

Background

graphical representation of Google Analytics account structure

In the Google Analytics hierarchy, you can alter the data set at the view level and by how the code is implemented.

Google Analytics follows a hierarchy of accounts, properties, views and reports. The account is the top-most level of organization.

Each account contains one or more properties. A property is associated with a unique tracking number. Within a property there are one or more views.

In the admin area you can change what data gets passed into the reports. For instance, at the view level you can add filters and set up goals and funnels. At the property level, you can set custom dimensions, which also require a code change.

Within the reports there are various ways to change how the data is reported that don’t alter the data itself. This includes segments, report filters and custom reports.

The data set is determined by how the code is implemented and by changes made at the view level. At the report level, you can manipulate how the data is displayed but you can’t change the underlying data set.

It is important to understand how changes at different points in the hierarchy impact the data because they have a profound effect upon the accuracy of your reports.

Determining Data

Tracking Code

Google Analytics assigns a unique code to each property. That code goes on the pages you wish to track if you are adding it directly to the site or you can add it using Google Tag Manager. Learn more about how to set up the tracking code.

Usually you will want to put the code on every page of your website. If you miss out pages then Google Analytics will not be able to track those pages. This sounds like an inconvenience but it can be fairly disastrous for your data.

For instance, when your visitors move from a tracked page to a non-tracked page it looks like they’ve left the site. If they then move back to a tracked page, they will appear as a new visitor. This will impact a range of metrics including the number of sessions, average time on site and bounce rate.

In the vast majority of cases, you’ll want to put your code on every page your visitors can access. If you don’t put it on every page, carefully consider how it will impact your data and how you can mitigate the results.

Filters

Within each property you can set up multiple views, each with different filters.

The best practice is to have one view which has no filters. This is because filters permanently alter your data. If you have a question that the filtered data can’t answer or there is an error in the filters, then you’ll need to go back to the view which has no filters.

I also recommend having at least one view for testing filters. This is to ensure that a new filter doesn’t accidentally destroy your data as the data is not recoverable. By running the new filter on a test view first you can make sure it does what you expect it to do.

Common filters include removing internal IP addresses, passing in sub-domains, or looking at sessions from specific sources.

Goals and funnels

At the view level you also set up goals and funnels.

A goal is used to track when a user completes an action, e.g. tracking when they visit a page or trigger a specified event. When setting up the goal you will also have the option of adding a funnel. A funnels tracks user behavior through a specific set of steps that lead to a goal.

A very common use case for funnels and goals is to track user behavior through a purchase flow where the last page of the purchase flow is the goal. The intent is to find out where users are dropping off during the purchase flow.

The goal and funnel information can be seen in the Google Analytics report associated with that view. Goals and funnels are not destructive. If you set up a goal or a funnel incorrectly, it won’t damage your underlying data but it may pass unusable data into your report.

eCommerce conversions

To enable eCommerce conversion tracking on your site you can use Google Tag Manager or add code to your conversion page, e.g. the last page of the purchase flow.

When conversion tracking is implemented correctly, Google Analytics will pass purchase information into your report. eCommerce tracking is extremely useful for tying site behavior to purchase information.

As with goals, if the eCommerce tracking is set up incorrectly, it won’t damage the rest of your data but it may pass in bad information.

Conclusion

The data that gets passed into your reports is determined at the view level and by how your code is implemented. If a mistake is made at this level you can’t recover the data.

In this post I focused on some of the most common ways to alter your data set. There are others, though, that I will explore in more detail in later posts.

In the next post, I will look at some of the ways you can work with data at the report level.

Update: Read the related post Google Analytics: Working with the data set.