# Likert Scales and summed scores

I cannot seem to find consensus on the following so hopefully someone can shed some light on it. I have 5-point Likert scale Strongly Agree, Agree, Neither Agree nor disagree, disagree, strongly disagree.

My question is measuring privacy attitude using a construct that contains 10 privacy indicators (researched separately). I am trying to create a privacy score so I can later do a correlation analysis - those who have a higher privacy score are more likely going to have a higher privacy activism score (another construct).

Without going into to much detail I can't seem to figure out the correct way to do the score. The most common approach seems to be a sum all the items in the construct (measuring privacy). However I am concerned about the undecided (neither agree nor disagree) group.

How should I rank that subset of responses. Logic would have it to be coded as 0 given they have not really answered the question? other literature suggests it should be 3.

My concern is, if (hypothetically) I have 30% of undecided and 20% strongly agree or agree, then the privacy score isn't really reflecting a persons desire for privacy because the undecided is scored at 3, pushing up the score.

SO the question is what to score the undecided group so I can create a summed score that accurately represents the response

thus, 1 = Strongly Agree
2 - Agree
3 = undecided
4 - disagree
5 - Strongly disagree

I would be grateful for any literature or suggestions I could reads regarding this.

• Likert data are ordinal categorical. For reasons such as you mention in your question, treating Likert data as interval numerical (so that means can be computed) is, and IMJO deserves to be, controversial. Using sample medians might be useful. Commented Dec 9, 2021 at 6:47

(more of a long comment)

How should I rank that subset of responses. Logic would have it to be coded as 0 given they have not really answered the question? other literature suggests it should be 3.

I guess we would need to know exactly how the questions was asked/worded. But I do not understand how you want to code undecided as zero, if it really is a non-answer it would be closer to missing? A related post is Ordinal variable with 'don't know'

Then, how do you make a score, summated or otherwise? Maybe first look at how the ten likert variables are correlated, maybe some are almost versions of each other? They would then get too much influence in a summated score ... have a look at How to choose between Pearson and Spearman correlation? and Under what conditions should Likert scales be used as ordinal or interval data?.

Then you could do an ordinal-data version of correspondence analysis, see Can principal component analysis be applied to datasets containing a mix of continuous and categorical variables? and How to understand optimal Scaling in R: The Package homals for novices. The scores on the first axis could be a candidate for a score, but I would not be surprised if that is highly correlated with the sum score.

• Thank you I have followed up on the above and cannot find a definitive answer. Likert himself suggested 1,2,3,4,5 - thus summing these score would give me an idea of which way the set is skewed . The highest score being 50 and lowest being 10 with battery of 10 items. I do like the idea of weighting the items but i could no find enough detail on how to do this. You are certainly correct in assuming that similar items could add to much weight. I will continue looking. re the scale score, I was also thinking about -1, -2, 0, 1, 2 , yet to find literature to support it.
– Leah
Commented Dec 10, 2021 at 8:26

Thanks to all who viewed or answered my post. For those struggling with the concept of aggregated or summed scales I have found some very good resources, that perhaps i should have found earlier, nonetheless it might help someone so posted here. Most obvious from Likert, The Method of Constructing a Likert scale, provided the basics of scaling. The most useful however is an article, How to Analyze Likert and other rating scale data. Finally, Summmated rating scale construction is a quick guide to defining and constructing scales. I am yet to discover the best way to weight each item. Some literature has touched on weighted factor analysis to do this however i need more reading time. The take away from the literature this far is understanding the difference between a Likert item and a Likert scale, what the two primary ways of scoring items, i.e., sum of means or sum of observed scores. I hope this is of help to the community.