Analyzing survey data is an important and exciting step in the survey process.
It is the time that you may reveal important facts about your customers,
uncover trends that you might not otherwise have known existed, or provide
irrefutable facts to support your plans. By doing in-depth data comparisons,
you can begin to identify relationships between various data that will help you
understand more about your respondents, and guide you towards better decisions.
This article gives you a brief overview of how to analyze survey results. It
does not discusses specific usage of eSurveysPro for conducting analysis as it
is intended to provide a foundation upon which you can confidently conduct your
own survey analysis no matter what tool you use.
Three Common Mistakes
Before you dive into analyzing your survey results, take a look back at the big
picture. What objectives were you trying to accomplish when you created your
survey? Did your survey instrument meet those objectives? Is the data you
collected the right data? Do you have sufficient data to properly reach a
Although data analysis is the wrong time to try and rewrite your survey
instrument, it is important to remember the scope of your project and stick to
it. Many first time surveyors attempt to read "between the lines" while
analyzing data. They attempt to answer questions that were not asked by making
inferences and assumptions from those that were asked. Doing so amounts to
nothing more than guesswork. To avoid this temptation, remember this simple
Rule 1: If you did not ask you do not know.
Another common mistake that many first time surveyors make is to attempt to
change data to compensate for poor question design. For example, if a question
asked a respondent to indicate his total household income using a scale of
values, a mean and median cannot be calculated. Many people try to get around
this by assigning each response a value representing the range. Even if the
adjustment is made consistently across all responses, the resulting
calculations will be wrong. Similarly, trying to analyze a multiple-choice
question as if it was a single-select question will often provide erroneous
information. In order to avoid this pitfall, remember this simple rule:
Rule 2: Do not alter data to compensate for bad survey design.
A second mistake inexperienced surveyors make is to project the findings to an
audience that was not either part of the survey population or not adequately
represented. For example, if an HR manager conducts a benefits survey and
invites all employees to participate, most people would assume that the results
represent all employees since everyone had an opportunity to participate.
Provided that enough employees participate, the data might be statistically
valid, but is it really representative of all employees? The answer is, it
depends. If the survey collected data about employee demographics that could be
compared to what is known about the company, then the results do reflect the
company as a whole. However, if 80% of the respondents are married and 50% of
the total employee base is married, the results of the survey are skewed toward
married people. If married people have different benefits needs than single
people, using the survey results to make conclusions about the entire employee
pool would be less accurate than those conclusions about the married employees
or single employees independently. To avoid this temptation, remember this
Rule 3: Do not project your data to people that did not respond.
The earlier you recognize flaws in your survey design and data collection, the
more time you will save during analysis. If you questions do not provided the
data you need to meet your survey objectives, you'll have to start over. If
your questions are vague or ambiguous, you'll have to throw them out. If you do
not have an adequate number of responses, you'll have to get more.
Analyzing any survey, web or traditional, consists of a number of interrelated
processes that are intended to summarize, arrange, and transform data into
information. If your survey objective was simply to collect data for your
database or data warehouse, you do not have to do any analysis of the data. On
the other hand, if your objective was to understand the characteristics of
typical customers, then you must transform you raw results in to information
that will enable you to paint a clear picture of your customers.
Assuming you need to analyze the data collected from your survey, the process
begins with a quick review of the results, followed by editing, analysis, and
reporting. To ensure you have accurate data before investing significant time
in analysis, it is important that you do not begin analyzing results until you
have completed the review and editing process.
Read all your results. Although, this seems like an obvious thing to do, many
surveyors think that they can skip this step and dive right in to data
analysis. A quick review can tell you lots about your project, including any
flaws in questionnaire design or response population, before you spend hours of
time in analyzing the data.
During the quick review, you should look at every question and see if the
results "make sense". This "gut feel" check of the data will often uncover any
issues with your survey project. Most surveyors already have an idea of how
they expect their data to look. A quick review of the data can help you quickly
understand that tell you if the people that respond are the right people. For
example, if you were conducting a survey of all the employees in a company and
you knew that 10% were in the marketing department, 20% in sales, 45% in
manufacturing, 5% in management, and 5% finance, and 15% research and
development, you could reasonable expect your responses to be similarly
distributed. If your quick review disclosed 80% of your respondents were from
the sales department, you know that your survey did not adequately capture a
representative sample of all departments within the company.
The quick review can also highlight any problems with the survey instrument. Are
most respondents answering all questions? If not, your questionnaire could be
flawed in such a way that a person cannot complete the survey. A low response
rate could mean your survey invitation was not compelling enough to encourage
participation, or your timing was off and a follow-up reminder is needed.
Lastly, the quick review of the survey can show you what areas to focus on for
detailed analysis. As stated earlier, most surveyors already know what they
expect to get, so your quick review can show you the unexpected.
Editing and Cleaning
Editing and cleaning data is an important step in the survey process. Special
care must be taken when editing survey data so that you do not alter or throw
out responses in such a way as to bias your results. Although you can begin
editing and cleaning your data as soon as results are received, caution should
be used since any edits can be lost if the database is rebuilt. To be safe,
wait until all data is received before you begin the editing and cleaning
To start, find and delete incomplete and duplicate responses. A response should
be discarded if the respondent did not complete enough of the survey to be
meaningful. For example, if a your survey was intended to determine future
buying intentions across various demographic groups and the respondent did not
answer any of the demographic questions, you should delete the response. On the
other hand, if the respondent answered all the demographic questions but
omitted their name or email address, then you should keep the response.
Duplicate responses are a unique issue for electronic surveys. Many tools, such
as eSurveysPro, provide built in features to help minimize the risk of
duplicate responses. Others, like the popular "infotainment" polls featured on
many websites do nothing to eliminate duplicates. Without removing duplicates,
your data will be skewed in favor of the duplicate response. Both the count and
percentage of the whole will be affected by duplicate responses, and computed
means and medians will also be thrown off. To find duplicate responses,
carefully examine the answers to any open-ended questions. When two open-ended
questions have the exact same answer, a duplicate response is likely to exist.
Make sure the response is indeed a duplicate by comparing the answers to all
the other questions, and then delete one of the responses if a match is found.
Data cleaning of web surveys usually involves categorizing answers to open-ended
questions and multiple-choice questions that include an "other, please specify"
response. Because of their nature, open-ended text response questions can
provide significant value but they are nearly impossible to process without
some form of summarization or tabulation. One of the easiest ways to summarize
these questions is to build a list of themes and select the themes that apply
as you read each response. Tools such as eSurveysPro allow you to add questions
after a survey is run to do just this sort of thing.
A common problem in any survey that needs attention during the editing and
cleaning process is when a respondent answers an "other, please specify"
question by selecting "other" and then writing in an answer that was one of the
listed response options. Without cleaning these answers, the "other" response
will be overstated and the correct response will be understated. For example, a
demographics question that asks for the respondent's role within the
organization may have a response like "faculty, teacher, or student" and a
respondent selects "other" and types "professor," you would want to clean the
response by switching the other choice to the one for "faculty, teacher, or
Once the data preparation is complete, it is time to start analyzing the data
and turning it into actionable information.
Analysis is the most important aspect of your survey research project. At this
point, you have collected a set of data that must now be turned into actionable
information. The process of analysis can lead to a variety of alternative
courses of action. Mistakes during analysis can lead to costly decisions down
the road, so extreme caution and careful review must be followed throughout the
process. Carelessness during analysis can lead to disaster. What you do during
analysis will ultimately determine if your survey project is a successful or
Depending on what type of information you are trying to know about your
audience, you will have to decide what analysis makes sense. It can be as
simple as reviewing the graphs that eSurveysPro automatically creates, or
conducting in-depth comparisons between questions sets to identify trends or
relationships. For most surveyors, a basic analysis using charts, cross
tabulations, and filters is sufficient. On the other hand, more sophisticated
users may wish to do a more complex statistical analysis using high powered
analytical tools such as SPSS, Excel, or any number of number crunching
applications. For our purposes in this article, we will focus on basic analysis
Graphical analysis simply means displaying the data in a variety of visual
formats that make it easy to see patterns and identify differences among the
results set. There are many different graphing options available to display
data, the most common are Bar, Pie, and Line charts.
Bar charts use solid bars on an X and Y-axis that extend to meet a specific data
value indicated on the chart and can be shown either vertically or
horizontally. These charts are flexible and are most commonly used to display
data from multiple-select, rank order, single-select matrix and numerical
questions. Each response option is shown as an independent bar on the chart,
and the length of the bar represents the frequency the response was chosen
relative to all choices.
Pie charts, or circle graphs, have colorful "slices" representing segments of
your data. These charts measure values as compared to a "whole", and the total
percentages of the segments always add up to 100%. Pie charts are most useful
with single-select questions because the each response is represented visually
as a portion of the entire pie. It is easy to interpret which answer received
the most responses in a pie chart by selecting the largest potion of the pie.
When comparing two sets of data using a pie chart, it is important to make sure
the colors used for each response option remain consistent in each chart. If
represent the same response options in each chart, this way, a side-by-side
visual comparison can quickly be made. Pie charts are not appropriate for
multiple-select questions because each respondent can answer choose more than
one option, and the sum of the option percentages will exceed 100%.
There are other graphing options such as line charts, area charts and scatter
graphs, which are useful when displaying the same data over a period of time.
However these formats are not as easy to interpret for casual users, so they
should be used sparingly.
Frequency tables are another form of basic analysis. These tables show the
possible responses, the total number of respondents for each part, and the
percentages of respondents who selected each answer. Frequency tables are
useful when a large number of response options are available, or the
differences between the percentages of each option are small. In most cases,
pie or bar charts are easier to work with than frequency tables.
|Creating sales tools
|Providing channel support
Cross tabulations, or cross tabs, are a good way to compare two subgroups of
information. Cross tabs allow you to compare data from two questions to
determine if there is a relationship between them. Like frequency tables, cross
tabs appear as a table of data showing answers to one question as a series of
rows and answers to another question as a series of columns.
|Product Marketing Manager
|Technical Product Manager
Cross tabs are used most frequently to look at answers to a question among
various demographic groups. The intersections of the various columns and rows,
commonly called cells, are the percentages of people who answered each of the
responses. In the example above, females and males had relatively similar
distribution among various job titles, with the exception of the tile of
"Technical Product Manager", where 2.5 times as many males had the title as
compared to females. For analysis purposes, cross tabs are a great way to do
Filtering is the most under-utilized tool used in analysis. Filters allow you
select specific subsets of data to view. Unlike a cross tab, that compares two
questions, a filter will allow you to examine all questions for a particular
subset of the responses. By viewing only the data from the people who responded
negatively, look at how they answered other questions. Find patterns or trends
that help define why a person answered the way they did. You can even filter on
multiple questions and criteria to do a more detailed search if necessary. For
example, if you wanted to know the buying intentions of men, over the age of
40, with income of about $50,000, you would set a filter that would remove all
those respondents that do not meet your criteria from the results set, thus
enabling you to concentrate on the target population.
By applying filters to the date survey responses were received, you can see how
the answers change from one time frame to the next. For instance, by
continually running a customer satisfaction survey, you can assess changes in
customer attitudes over time by filtering on the date the survey was received.
You can also use a filter on date received to assess the impact of sales
incentive programs or new product offerings by comparing survey responses
before and after the change.
Filters do not permanently remove the responses of those people that do not
match the specified criteria; they simply eliminate them from the current view
of the data, making it much easier to perform analysis. By looking at the same
question with different filters applied, differences between the various
respondents represented by the filter can be quickly seen. Because filters
remain in effect until cleared, don't forget to clear them before attempting to
analyze your survey responses as a whole, otherwise your observations will be
inaccurate, and your recommendations flawed.
Simple Regression Analysis
Determining what factors have lead to a particular outcome is called regression
analysis. The regression means you're working backwards from the result to find
out why a person answered the way that they did. This can be based on how they
answered other questions as well.
For example, you might believe that website visitors who had trouble navigating
within your website are likely not return again. If 30% of the respondents said
they had trouble navigating through the website and 40% said they would not
return, you could look at only those that would not return to determine if poor
navigation might be the case. After filtering to only those who would not
return, if 30% or less said they had trouble navigating, then this is clearly
not the "reason" visitors will not return. By filtering out those that would
return, we expect the percentage to increase dramatically. If it does, we still
cannot conclude that navigation is "the" reason, only that it might contribute
to the respondents not returning. In order to know if it is "the" reason, we
would need to ask a direct question.
After analyzing your survey data, it is time to create a report of your
findings. The complexity and detail need to support you conclusions, along with
your intended audience, will dictate the format of your report. CEO's require a
different level of detail than line managers, so for maximum results consider
who is going to receive your report and tailor it to meet their unique needs.
Visual reports, such as an HTML document or Microsoft PowerPoint presentation,
are best suited for simple findings. These graphical reports are best when they
are light on text and heavy on graphs and charts. They are reviewed quickly
rather than studied at length, and most conclusions are obvious, so detailed
explanations are seldom required. For more complex topics, a detailed report
created in Microsoft Word or Adobe Acrobat is often required. Reports created
using Word often include much more detailed information, report findings that
require significant explanation, are extremely text heavy, and are often
studied at great length and in significant detail.
No matter which type of report you use, always remember that information can be
more powerfully displayed in a graphic format verses a text or tabular
representation. Often, trends and patterns are more obvious and recommendations
more effective when presented visually. Ideally, when making comparisons one or
more groups of respondents, it is best to show a chart of each group's
responses side-by-side. This side-by-side comparison allows your audience to
quickly see the differences you are highlighting and will lead to more support
for your conclusions.
At the beginning of your report, you should review your survey objective and
sampling method. This will help your audience understand what the survey was
about, and enable you to avoid many questions that are outside of your original
objectives. Your report should have a description of your sampling method,
including who was invited to participate, over what time frame results were
collected, and any issues that might exist relative to your respondent pool.
Next, you should include your analysis and conclusions in adequate detail to
meet the needs of your audience. Include a table or graph for each area of
interest and explain why it is noteworthy. After your analysis section, you
should make recommendations that relate back to your survey objectives.
Recommendations can be as simple as conduct further studies to a major shift in
company direction. In either case, your recommendation must be within the scope
of your survey objective and supported by the data collected. Finally, you can
include a copy of your survey questions and a summary of all the data collected
as an appendix to your report.
Survey analysis is not as easy as downloading results and printing a chart or
report, yet it is not so complex that it requires a PhD. In this article we
have learned that good analysis begins with good questions, representative
participation, and careful interpretation of the data, in order to produce
actionable results. Techniques such as charting, filtering, cross tabulation,
and regression analysis all help you spot trends and patterns within your data
while helping you meet your survey objective. You now have a solid foundation
upon which you can confidently conduct your own survey analysis using a tool