Likert Scale Analysis - Newb Needs Help!

T

timmywimmy

Hi guys,

I'm a complete Minitab newb!

I've devised and completed a Likert Scale for a group's opinions and can't find out how to enter it into and analyse it with Minitab.

Help!

Cheers, Tim
 
B

Barbara B

There are different ways in which you can go on, but let me first ask you some questions:
What format does your data have (xls? txt? paper?)
Have you already decided which statistical method you want to conduct (e. g. metrics/descriptive statistics, tests)? (If so, please name them.)
What is the purpose of your analysis, for example which question should be answered based on your data?

Regards,

Barbara
 
T

timmywimmy

Hi Barbara B,

Thanks for your interest!

I've inputted the data manually and it is the results of a number of questions. I'm sorry, I can't reduce the text size to line it all up properly, but it is a five choice Likert scale, with 12 questions. the numbers are the number of votes each question had for each choice.

Question Strongly disagree Disagree Neither agree nor disagree Agree Strongly agree
I know what the team mission statement is 1 3 2 3 0
I know what my specific role is 0 0 1 4 4
I know what the team mission is 0 2 3 4 0
I know my role in the overall team mission 0 0 0 7 2
I think the team mission is going well 1 4 4 0 0
I think the team mission is worth the money 1 3 4 1 0
I know what the future holds for the team mission 1 4 4 0 0
I know what my specific role is in future development 0 1 3 3 2
I think enough effort is put into developing the team goals 0 1 6 1 1
I think enough is being done to develop the team's role for the future 0 5 2 0 0
I think the team is improving ability to patrol 1 4 2 2 0
I think the team mission is worth it overall 1 1 3 3 1

Any ideas gratefully appreciated!

As for which statistical analysis I want to do, I want to summarise the team's feelings before a project (data shown above) and then summarise the team's feelings after a project, following re-issue of the above Likert scale to complete again - thus I can see if there have been any improvements in opinion/attitude.

Tim
 
B

Barbara B

My recommendation is to build a table in which the ratings (not the number of votings per category) of every question are assigned to one column. Attached you'll find an excel file with two worksheets, "frequencies" with number of votings per category and "raw data" with the "original" ratings (values "after" made by hand.)

One method to describe ordinal data like in a Likert scale is to use boxplots. You can compare the ratings before and after for each question seperately or for all questions in one chart (see pdf attached).

Hope this helps,

Barbara
 

Attachments

  • Likert Scale 2011 01 28.xls
    21.5 KB · Views: 377
  • Likert Scale 2011 01 28.pdf
    488.9 KB · Views: 377
T

timmywimmy

Wow Barbara - thanks so much! This looks like it might be just what I need. I'll have a play and let you know how it goes - thanks so much! Tim
 
T

timmywimmy

I managed to get some very nice boxplots, thanks (would another useful way of doing it be by using interval plots - a bit simpler looking?).

Overall, I can see an improvment in most of the questions posed in the Likert survey. The question is, is there a next step, or one that most people go to?

Perhaps to:

1. Quantify the percentage improvements overall?
2. Quantify the percentage improvement per question?
3. Study interactions between question results.

etc!

Very grateful for any thoughts!

Regards, Tim
 
B

Barbara B

Tim,

interval plots are drawn based on the mean and the confidence interval of the mean or the standard deviation.

For an ordinal scal like the likert scale with not necessarily equally spaced values the mean isn't a good choice to characterize your data, so boxplots are preferred for ordinal measurements as they are simply based on robust metrics like median and quartiles.

To take the analysis one step further you can compare the medians (of each question or overall) previous vs. after, first by simply writing them in a table. Furthermore you can test with the so called nonparametric methods if the medians previous and after differ significantly:

Stat > Nonparametrics > Mann-Whitney
First Sample: [1 column with ratings previous]
Second Sample: [1 column with ratings after]

H0: "medians are the same" / median_previous = median_after (In Minitab: Alternative "not equal")
p-value < 0.05: medians differ significantly
p-value >=0.05: no significant difference between the medians previous and after

In the Minitab Session window you can find the p-value in the last sentence of the output
"The test is significant at 0.2679 (adjusted for ties)" with 0.2679=p and "adjusted for ties" meaning that the test statistic is corrected in case identical ratings (like 4 times "5: strongly agree") occur.

If you want to compare all ratings previous and after you have to rearrange your data in a way that you have one column with the "previous" and a second column with the "after" ratings.

You can also compare more than two medians, e.g. to test if at last one median differs from the others significantly. Therefore all ratings have to be copied in one column with a second column in which the group or category from a median is given:
Rating Group
2 Q1 prev
1 Q1 prev
4 Q1 prev
2 Q1 prev
3 Q1 aft
4 Q1 aft
5 Q1 aft
5 Q1 aft
2 Q1 aft
3 Q2 prev
1 Q2 prev
2 Q2 prev
4 Q2 aft
5 Q2 aft
2 Q2 aft
3 Q2 aft
...

The method for testing 2 or more medians if the are equal is the "Kruskal-Wallis-Test" which you can find in Minitab in the following menu:
Stat > Nonparametrics > Mann-Whitney

The Kruskal-Wallis-Test is the same test like the Mann-Whitney-Test if only 2 groups are analyzed.

Quantifying your test results is a little bit stretching the study, because as stated earlier you do not have equally spaced values. In the subjectivity of each study participant there is not necessarily the same distance between e.g. "1: strongly disagree" - "2: disagree" and "2: disagree" - "3: neither agree nor disagree".

So the categorizing with numbers leads to an ordinal scale and is therefore a wobbly basis to calculate percentages as they are equally spaced: difference between 10% and 20% is the same as the difference between 20% and 30%.

Regards,

Barbara
 
T

timmywimmy

Thanks Barbara - will be giving this one a go!

Kind regards,

Tim
 
S

Stats_Novice

Hello Barbara,
I am trying to analyse Likert scale data for my research too. Just like timmywimmy here, I am a complete newbie to minitab. Can you please help me with my analysis? I have created a very simple questionnaire with 5 questions. The answers to each of those questions are in the form of a 4-point likert scale. Can you advice the best way to analyse this data? I have read so many things on the analysis of likert scale data that I am thoroughly confused now. To give you an example, I have
Q.1 The essence of financial regulation has no strategic impact on an organization
1. Completely agree; 2. Slightly agree; 3. Slightly disagree; 4. Completely disagree.

I have a very small sample size (about 20) as I am trying to get the opinions of middle and upper level bankers on financial regulatory reforms. On getting the answers, I want to use the question statement into a null hypothesis and accept or reject the hypothesis based on my findings. I have downloaded minitab too. I'd be extremely grateful if you could you please tell me what would be the best way to do this?

Thanks a ton.
 
Top Bottom