Analyzing Survey Results

One aspect of market research is analyzing thethat the bunching of responses among a few
results of questionnaires and other surveychoices might create difficulties in ranking
"instruments." There are numerous analytical testsresponses. This is especially true if those thousand
for this purpose and most are classified asare divided into several cohorts.
"nonparametric" tests.The latter point is why this writer usually
"Nonwhat!?"recommends three or five choices per question
Don't sweat it. This article focuses on more basicwhen the object is one of ratings or satisfaction.
analysis that can be quite useful and that mostThe choices should be balanced. That is, there
people can perform fairly easily, especially withshould be the same number above a neutral or
computer spreadsheets."average" response as there are below. Also the
Before performing any type of analysis, the datachoices should be comparable in degrees from
must be tabulated. Tabulating means totalingthat middle or neutral or average point. For
(adding) the number of responses to each choiceexample, choices might be "excellent," "good,"
for each question. For instance, if there are 10"fair," "poor," and "unacceptable." (A,B,C,D,E). Or
questions, each with five choices, there are 50"good to very good," "fair or average," and "poor
different totals. Also, there are techniques forto unacceptable."
coding (giving a quantitative scale to) open-endedE and "unacceptable" are being used to avoid this
(qualitative) questions.being too much like a school report card.
Almost always the first step is to determine whatNOTE: With average, use "average" not "just
percent each choice is for each question. Generallyaverage." The "just" tends to denigrate the choice
this means, for example, calculating what percentof average. Kind of like hosts or hostesses at
of question 1's answers were choice A, whatrestaurants who say "just one" instead of "one."
percent were choice B, what percent were choiceRemember, by definition in a bell curve (i.e.;
C, what percent were choice D, and whatnormally shaped statistical distribution) average is
percent were choice E. This is repeated for eachthe middle 68+ percent. So when some of these
question. Thus, if you sent out 10,000 surveyspolitical and other polls' reports say positive is
and received 200 valid replies, that is a two"good or excellent" and negative is "just fair or
percent response rate, which for a generalpoor" that is really misleading. Fair is average. It is
audience survey is not too bad. It is possible notnot positive or negative. That is why a balanced
every question will produce 200 responses. Onescale with an odd number of choices is preferable
of the advantages in using percents rather thanfor an accurate reading.
absolute numbers is that most problems arisingDepending on the objective(s) of the survey,
from some respondents not answering eachnormally one would rank all the A responses, B
question are negated.responses, C responses, D responses, and E
The survey should try to gather key cohortresponses. It really doesn't matter what number
information. A cohort is a grouping, such asscale is used as long as the scale is symmetrical.
gender, age bracket, income bracket, etc.For example,
You can then look at what percent of eachA=1
cohort gave what answers. That helps develop aB=2
profile.C=3
NOTE: It is often easier to get answers andD=4
certainly easier to tabulated responses if you useE=5
if you "bracket" or "group" some answer ranges.Of course the scaling can just as easily go the
Grouping or bracketing also helps encourageother way. It's a matter of personal preference.
responses for sensitive information. TheseHowever, in that case it's worth noting that using
sensitive areas include age and income. What thezero instead of one for the lowest scoring choice
brackets should be depends on what you arecould cause some difficulties with some
trying to determine.quantitative tests.
In designing surveys, it is helpful to keep thisRegardless, the analysis will probably determine
tabulating and analyzing process in mind. Twothe arithmetic mean, the median, and the mode
particular considerations are:for each question and for the survey overall.
A) Whether the responses are not "computerThen the same thing can be done by cohort and
graded," that is read and tabulated by a computerby multiple cohorts. For example, the tabulation
(or earlier machine.andand cross-tabulation can show the results for
B) The more choices there are per question, thewomen, for women in their 40s, for women in
less the typical number of each response pertheir 40s with a household income in the
question. This sometimes makes it difficult to$35K-$50K range.
determine differences in responses. For instance,Numerous comparisons are possible depending on
if you have 1,000 responses, which is normallyjust what you want and how much time or
plenty, divided among 10 choices, it's quite possiblemoney you want to spend on this project.