Likert Scale Computation

sndz

New Member
Hello,

i have a likert scale questionnaire where respondents will ranked questions 1(Strongly Agree)-6(Disagree). the research is about comparing two software applications on which users prefer using.

Problem is how will i compute for the data and how will i compare them. i have only tallied the responses and that's where i left off. I'm so lost.

Any help will be appreciated.THANKS

trinker

ggplot2orBust
How many questions and are they all measuring the same construct (like of software)?

sndz

New Member
How many questions and are they all measuring the same construct (like of software)?
there are 37 questions and 50 respondents. i'm trying to comapre wordpress and joomla. the questions are basically about this two software like for ex. one of the questions would be if the application is easy to install or is a dynamic software, etc.

so basically base from the data that i gathered 33 pick wordpress and 17 pick joomla.

P

parsec2011

Guest
There are three main considerations here. First, it looks like your data was not gathered from a random sample of software users. Second, even if the data was gathered randomly, in this case, you are dealing with a rather small sample size. Last, statisticians commonly warn against using parametric testing (hypothesis testing methods such as t-tests and the like) because Likert-scales are a sort of "somewhere in-the-middle" concept between ordinal and interval data. In other words, the boundary between the “good” item/response on the one hand and a “very good” item on the other is hard, if not impossible to establish making full mathematical sense. Because of this, it can be questioned whether using inferential statistics (t-test) to compare your results is a categorically correct way to proceed.

That being said, what might be a more adequate solution could be to use the more common descriptive statistics concepts (mean or average and median) and to make a contingency table that includes two columns (one for wordpress and another for joomla) and five rows representing each of the Likert-scale items. Using the median, specifically is strongly advisable because an average or mean does not always accurately represent the value that is more common in the sample. This is a common problem across many small and large samples. The Excel functions =AVERAGE() and =MEDIAN() could help you to do this easily. Alternatively, you can use a calculator and track the list manually. Please refer to the article (1) for a more robust explanation and example of how this is done. You need to code your very good, good, average, etc. values into numbers (check process below) before calculating their average and median, and I recommend using a 0-10 scale (See process below).

If your data was taken randomly, you might want to use inferential statistics as a reassurance that the difference in perceptions between the two products –represented by the two mean scores- is statistically significant. In other words, a t-test using the two average values would help you to confirm with statistical certainty that users/clients prefer wordpress to joomla by conducting a test that takes into consideration the size of your sample, and the dispersion of the answers, both of which are crucial factors to make valid inferences on data.

In another thread in the forum (2), it is suggested that you can use a t-test to see if the mean response between two questions is different. To do this, you need to get the mean response for wordpress and joomla by calculating the average for each program respectively. First, you need to code your Likert-scale values into numeric data by assigning a number to each of the responses, and then, calculate the mean or average for each software program. A possible numeric scale is: Strongly agree (10), agree (7.5), neither agree nor disagree (5), disagree (2.5), and strongly disagree (0). Once you have the mean values for both, perform the t-test. The easiest way to accomplish this is by using an online calculator for t-test comparing averages or means. The best reference I have is (3). Notice you will need the standard deviation for both the wordpress and joomla datasets, which you can get from excel using the formula =STDEV().

Again, you need random data for this procedure; larger samples yield results that are more accurate. A description of t-tests in this context is provided in full at (4).

I have experience with qualitative research and college level statistics and this problem is not within the common list of statistics problems. Therefore, I advice you to look for more opinions concerning this research issue.

All the best

-------------

References:

(1) http://asq.org/quality-progress/2007/07/statistics/likert-scales-and-data-analyses.html

(4)http://biology.nebrwesleyan.edu/Courses/Labs/Biology_of_Animals/Statistics_Ttest.html

sndz

New Member
There are three main considerations here. First, it looks like your data was not gathered from a random sample of software users. Second, even if the data was gathered randomly, in this case, you are dealing with a rather small sample size. Last, statisticians commonly warn against using parametric testing (hypothesis testing methods such as t-tests and the like) because Likert-scales are a sort of "somewhere in-the-middle" concept between ordinal and interval data. In other words, the boundary between the “good” item/response on the one hand and a “very good” item on the other is hard, if not impossible to establish making full mathematical sense. Because of this, it can be questioned whether using inferential statistics (t-test) to compare your results is a categorically correct way to proceed.

That being said, what might be a more adequate solution could be to use the more common descriptive statistics concepts (mean or average and median) and to make a contingency table that includes two columns (one for wordpress and another for joomla) and five rows representing each of the Likert-scale items. Using the median, specifically is strongly advisable because an average or mean does not always accurately represent the value that is more common in the sample. This is a common problem across many small and large samples. The Excel functions =AVERAGE() and =MEDIAN() could help you to do this easily. Alternatively, you can use a calculator and track the list manually. Please refer to the article (1) for a more robust explanation and example of how this is done. You need to code your very good, good, average, etc. values into numbers (check process below) before calculating their average and median, and I recommend using a 0-10 scale (See process below).

If your data was taken randomly, you might want to use inferential statistics as a reassurance that the difference in perceptions between the two products –represented by the two mean scores- is statistically significant. In other words, a t-test using the two average values would help you to confirm with statistical certainty that users/clients prefer wordpress to joomla by conducting a test that takes into consideration the size of your sample, and the dispersion of the answers, both of which are crucial factors to make valid inferences on data.

In another thread in the forum (2), it is suggested that you can use a t-test to see if the mean response between two questions is different. To do this, you need to get the mean response for wordpress and joomla by calculating the average for each program respectively. First, you need to code your Likert-scale values into numeric data by assigning a number to each of the responses, and then, calculate the mean or average for each software program. A possible numeric scale is: Strongly agree (10), agree (7.5), neither agree nor disagree (5), disagree (2.5), and strongly disagree (0). Once you have the mean values for both, perform the t-test. The easiest way to accomplish this is by using an online calculator for t-test comparing averages or means. The best reference I have is (3). Notice you will need the standard deviation for both the wordpress and joomla datasets, which you can get from excel using the formula =STDEV().

Again, you need random data for this procedure; larger samples yield results that are more accurate. A description of t-tests in this context is provided in full at (4).

I have experience with qualitative research and college level statistics and this problem is not within the common list of statistics problems. Therefore, I advice you to look for more opinions concerning this research issue.

All the best

-------------

References:

(1) http://asq.org/quality-progress/2007/07/statistics/likert-scales-and-data-analyses.html