Home › Forums › Default Forum › how to interpret likert scale data for correlation analysis
- This topic has 10 replies, 8 voices, and was last updated 10 years, 3 months ago by
Professor Paul Jackson.
-
AuthorPosts
-
24th September 2009 at 12:13 pm #5562
jaya
Membercan any one help me …….
i have prepared likert scale questionnaire.i want to do pearsons correlation analysis to find the relationship between two variables.
pl z tell me how to take data’s from questionnaire (likert scale) to find correlation coefficient.
jaya
11th November 2009 at 9:43 am #5572Professor Paul Jackson
Membernot sure why no one has picked this up. A Likert scale can be considered as a grouped form of a continuous scale, and so you just treat the variable as if it were continuous for correlational analysis. Likert scales are clearly ordered category scales, as required for correlational work, and the debate aong methodologists is whether they can be treated as equal interval scales. My own view is that it makes very little difference provided that the data are distributed in a broadly symmetrical way along the scale.
I’ve discussed this more in my book – Management Research by Easterby-Smith, Thorpe and Jackson.
Regards
Professor Paul Jackson
Manchester Business School13th November 2009 at 6:27 pm #5571Jeremy Miles
ParticipantCan you be more specific. You enter the data into a computer, and then you correlate them. Which part is giving you trouble?
14th November 2009 at 8:57 am #5570K.Kalyanaraman
MemberThere are two questions relating to Likert Scale. They are available under links ‘Analysis of Likert scale Questions’ and ‘How to interpret Likert scale data for correlation analysis. The first one seeks to know the nature of Likert Scale and if they can be used for correlation and chi square test. The second one has a question regarding the data organization. I think there is a lot of confusion with the use of measurements using this scale.
Let me look into the problem from the beginning. Also, I shall try to add other required information purely for academic purpose.
Measurement in research consists in assigning numbers to entities otherwise called concepts in compliance with a set of rules. These concepts may be ‘physical’, ‘psychological’ and ‘social’. The concept ‘weight’ is physical and can be measured using any of the standard scales and report the weight in ‘pounds’ or ‘kilograms’. This does not pose any difficulty. However, measurement of ‘psychological’ and ‘social’ concepts may not be directly possible as there are no standard scales. Naturally, a scale has to be developed and ‘Likert scale’ is one of the widely used scaling methods for this purpose.
The ‘Likert Scale’ is an INTERVAL SCALE. Let us see how it can be considered as an Interval Scale.
Consider a concept that is physical in nature, say weight. Start with asking ‘what is meant by ‘Height’, i.e., by reporting height as 175 cm in a case, what exactly it means. One possible answer may be as the following:
‘There are many reasons for the concept ‘Height’ to be assigned a number 175 in a particular case. Factors related to ‘genetic’, ‘behavioral’, ‘food’ and other considerations, might be imagined to contribute to the concept ‘Height’ a share and the ‘Height’ is measured as the sum of these contributions as well as contributions from many other non-dominant factors. Note that the measure of the concept ‘Height’ is the SUM of the shares of the contributions of these different factors. It may not be difficult to see that this measure is under an ‘Interval Scale’ (Ratio scales are interval scales). The question is WHY? Height is a variable generally having a bell shaped distribution.At this point one has to address to Probability Theory. The genesis of a variable having a Normal distribution is very well known i.e. the value taken by such a variable is the sum of values contributed by a large number of factors related to the variable. (I am avoiding technical statements about this idea in probability Theory). Variable following a Normal distribution is expected to take any real value, thus turning out to be a measurement under ‘Interval Scale’.
Now look at the Likert Scale. The researcher uses the scale to measure abstract concepts by generating a number of statements and obtaining the responses in a, say, 5 point alternatives having an inherent order. Responses for positive statements are weighted with a decreasing set of equally spaced numbers, say 5,4,3,2 and 1. (Equally spaced numbers are used as the researcher feels that the five alternatives are equally spaced). Responses for negative statements are weighted in the reverse order. It is to be noted that the choice of these numbers are only for convenience. With the responses for each statement supposed to be related to the abstract concept being measured being assigned appropriate weights (measuring the contribution of a statement), the considered concept is measured using the SUM of all these weights. (Compare this with the narration associated with the physical concept of Height earlier). Thus the measurement under Likert Scale is under ‘Interval Scale’.
The next question of using them for analysis using correlation and Chi square, a simple answer is yes. However, be conscious of other principles associated with correlation. For using such measurements for Chi square (Note that there are a number of tests using Chi-square. Hope here ‘Test for Independence’ is what is intended), perhaps the data may have to be classified into an acceptable number of categories using some principle. For e.g. like measurements on ‘Motivation’ being used and classified into ‘High’, ‘Middle’ and ‘Low’ levels of motivation.
Interpretation of correlation will be in the usual way. Positive correlation between X and Y implies a possibility of a proportional increase or decrease in the value of X for an increase or decrease respectively in the value of Y and vice versa. Similarly, Negative correlation will imply a possibility of proportional decrease or increase in the value of X for an increase or decrease respectively in the value of Y and vice versa. Coefficient of correlation may be between -1 and +1. Nearer the correlation to +1 or -1 the possibility in the above statements are very high while it is low when correlation coefficient is nearer to zero.
Data on X and Y may be used in two adjacent columns in MsExcel.15th November 2009 at 2:52 pm #5569Kristian Karlson
MemberDear jaya,
I am always confused with the notion of “Likert scale”. I prefer using other labels such as “ordered discrete variable”, “variable measured on an ordinal scale”, or similar expressions. However, this is of minor importance.
Regarding correlation analysis: If you believe that your ordered discrete variables in fact are discrete approximations of underlying variables with “normal”, continous distributions, you might want to have a look at Stas Kolenikov’s Stata command -polychoric-: http://www.cpc.unc.edu/measure/publications/pdf/wp-04-85.pdf. The command enables you to estimate polychoric correlation on your “Likert type variables” without assuming true interval scale measurements.
Another option is to use Spearman’s rank order correlation (which is a nonparametric correlation coefficient). I recommend Wikipedia’s explanation of it:http://en.wikipedia.org/wiki/Spearman%27s_rank_correlation_coefficient.
All the best,
Kristian20th December 2009 at 8:00 am #5568jaya
MemberHai,
Thank you for your response.
MY PROJECT IS ROLE OF DYMNAMIC TECHNOLOGY ON THE LONG TERM SURVIVAL OF THE FIRM”HERE I LL WRITE SOME QUESTIONS IN THE QUESTIONNAIRE. MY SAMPLE SIZE IS 20.
I WANT TO CORRELATE FACTORS OF TECHNOLOGY MANAGEMENT WITH LONG TERM SURVIVAL.1.How often the deliveries of products are as per scheduled time?
Almost never
Rarely
Sometimes
Usually
Almost always2. How much do you agree your organization has provided between service and advanced design?
Strongly agree
Agree
Neither agree nor disagree
Disagree
Strongly disagree3. How often your organization is concentrating on new process and product development?
Almost never
Rarely
Sometimes
Usually
Almost always4. Technology promotion centers are established.
Yes
No5. How often your marketing team is provided with technical information to support post marketing services to the customers?
Almost never
Rarely
Sometimes
Usually6. To what extent do innovative inputs contribute to the insatiable appetite of the consumers for new and improved products?
Not at all
To a little extent
To some extent
To a considerable extent
To a very great extentSO TEL ME HOW TO TAKE DATAS FROM THE QUESTIONNAIRE (X&Y FOR CORRELATION.)
PLZ TEL ME CLEARLY AS SOON AS POSSIBLE.
8th September 2010 at 4:06 am #5567Vishwas Gupta
MemberHi to All,
Is there any weightage methods to caluclate the values of Likert Scale?
4th October 2010 at 5:43 pm #5566Chandan Rathore
ParticipantHI,
I would like to ask how to correlate the following variables Age, Gender, Ethnicity with attitude….I have used the Likerts scale with the options strongly agree, agree, disagree, strongly disagree……I have decided upon spearman’s rho however i have also come across multiple regression analysis….Could you please let me know what analysis on SPSS would be appropriate considering the type of variables i have used..
Thanks…waiting for your reply15th October 2010 at 7:39 am #5565K.Kalyanaraman
MemberDear Friend
You have variables attitude measured using a Likert Scale, age, gender and ethnicity. Note that Attitude and Age (if measured in completed years or months) are in interval scale, gender in nominal scale and I assume that ethnicity is also in nominal scale. In such a case you can think of the usual correlation coefficient (Pearson’s) bet ween attitude and age provided there is linearity between them. (verify using a scatter diagram). Correlation is not allowed for other cases, i.e. attitude vs ethnicity, attitude vs gender, age vs ethnicity or age vs gender. If you can group attitude scores into low, medium and high and transform them into ordinal scale then you can consider attitude (G), gender and ethnicity and find a measure of ASSOCIATION (which is similar to correlation). SPSS has this option under “crosstab”—> “statistics”.15th October 2010 at 5:33 pm #5564Sujeesh Kumar.S
MemberHi…
Please refer shri.kalyanaraman’s comment on your qustions. i am also suggesting the same. i would like to add the second part of your question..you can run a multiple regression after checking the linearity of the variables (age,Ehtiniciy,attitude),and can regress attitude over age,ethinicity,and gender being a dummy variable.you can also go for some non parametric methods …15th October 2010 at 9:43 pm #5563Chandan Rathore
ParticipantThank You so much.
-
AuthorPosts
- The forum ‘Default Forum’ is closed to new topics and replies.