Interval Data Likert Scale Analysis
HTML-код
- Опубликовано: 15 июл 2024
- Interval data is data that is measured on a scale where each point is an equal distance from another. Interval data includes LIkert scales where ratings like "extremely agree" to "extremely disagree" are used. This tutorial shows you how to analyze interval data.
You must have IBM's SPSS and Excel to complete this analysis.
![XO (Only If You Say Yes) - ENHYPEN エンハイプン 엔하이픈 [Music Bank] | KBS WORLD TV 240712](http://i.ytimg.com/vi/wBT1bKKwPTo/mqdefault.jpg)
Hello Jennifer! Thank you so much for your explanations, very useful!! I had so many questions before, and you have already answered a bunch of them. However, I still have a question that cannot find anywhere on the internet.
So, I made a questionnaire and I have both Likert scale and ordinal responses. To interpret that data seems straightforward after watching your videos. The problem is that, for my paper, it seems meaningless and too much effort to go one by one with each question (31 questions in total). Also, I added 6 factor variables that will allow me to make comparisons between the respondents. Lastly, the point of my questionnaire was to measure a unique construct. Thus, for example, 7 out of my 31 questions measure job satisfaction, using like scale responses. Then I figured that it would be great to use a composite variable, where I could get a unique conclusion of job satisfaction combining all those 7 questions. So far everything great. However, I am not 100% sure how to present this data. Okay, I can calculate the mean, median, I can even do a boxplot that shows where the data is concentrated, and I thought to do a boxplot with each factor variable, comparing each of them, but that took me nowhere. Then I finally thought to make a line chart, just like a function, which can connect the point I have. But now the scale on the x-axis completely lost its meaning. After making the composite variable (I just used the mean of the responses, well I also did the sum but I get to the same problem), the responses are not "strongly disagree" to "strongly agree" anymore. Thus, I have no idea have to give a sense to, for example, having 2.9 or 3.2 in my means., or even in the graphic interpretation of the charts that I was trying to draw.
Please advise me on the use of composite variables in the Likert scale and ordinal responses. Any book, paper, link, video, or your so well-detailed explanations, would be much of a help for me right now. Thank you so much!!! Bests, Álvaro
Hi Alvaro! Thank you for your question. These are great questions and I know of several resources that can help, but before I can send you in the right direction, I need a bit more information. First, what is the purpose of your survey? By this I mean, are you completing a master's thesis, a doctoral dissertation or are you doing this for business? Second, what field are you in (Psychology, business, etc.). Third, what is your research question that you are trying to answer? I ask these questions, because there are different ways you can analyze the data that you describe, some are much more rigorous than others. If you would like a more private discussion, then please feel free to email me at jennifer.taylor@tamucc.edu. Either way, I will gladly help steer you in the right direction and provide you with resources to help you with your analysis.
Thanks much appreciated
it's so perfect
Hi Jessica, thanks for creating the video! This is a problem that happens to me frequently, but when you dragged the Valid Percent formula (Column E) at 10:35 it replaced your previously correct Valid Percent calculation to reference the Missing and Total cells in column C.
Hi Jimmy! Thank you very much for identifying this error. I never even noticed that. It sounds like I need to either redo this video or find a way to highlight and explain this error on the RUclips video. Just as soon as I find some extra time I will correct this issue. Thank you again for your help! I greatly appreciate you bringing this to my attention. Have a great day!
Hi，you said that 5 Likert scales are interval, I'm quite confused because many researchers insist that likert scales are ordinal data
You are exactly correct, many researchers insist that likert scales are ordinal. But I am not one of those researchers. The reason being is that you cannot report means for ordinal data - it is meaning less because there are unequal intervals between the scales. However, with Likert scales, there is an equal distance in the scale (1,2,3,4,5). Which means you can report means and it has meaning. For example if you have a Likert scale for customer satisfaction and where 1=very unsatisfied, 2=somewhat unsatisfied, 3=neither satisfied nor unsatisfied, 4=somewhat satisfied, 5=very satisfied, to have a mean of 3.5 tells us that the majority of people are satisfied. And assuming that the standard deviations were similar, if we were to compare the satisfaction rate between two companies we could safely say that customers are less satisfied with a company that has a satisfaction rate of 3.4 compared to 3.6. The point being that the mean is useful data.
The reason there is an argument as to whether Likert Scales are ordinal or interval is that researchers say that the labels that respondents choose from (very unsatisfied to very satisfied) does not have a true measurable interval and that we are assuming that all people that select very satisfied are equally distant in their satisfaction from those people who select somewhat satisfied. This is a compelling argument, but in reality, how you view Likert scales depends on your field of research. In my field (Marketing) it is generally accepted that you must report the means of Likert Data (making it an interval scale). This is especially true if you are doing advanced analyses with the scales, such as Structural Equation Modeling, where you will need to report the Likert Scales means in the results section.
My advice to you is to follow the generally accepted stance towards Likert scales for your field and if they follow the stance that Likert data is an ordinal scale, that you can't report the means.
@@jennifertaylor3020 Thanks for your time and explanation. My supervisor provided me with a particular way to tackle such a debate: Rasch Modelling.
@@jennifertaylor3020 Thanks for the arguments, could you maybe provide any literature to back it up? I am writing a master thesis in the field of marketing and so far I only see papers which state that likert scale is ordinal :(
Can you share data and models?
Is likert scale measured as interval data? In interval data, we can find the difference between between two or more data or we can add two or more data. But in case of likert scale, Is it possible to sum or substrate the data? Such as 1 is strongly disagree, 2 is disagree, 3 is neutral, 4 agree and 5 is strongly agree. Can we substrate 5 and 4? Or add 3 and 5? Here, we have just assigned numbers as codes to text. So, now I am perplexed if likert scale shall be treated as ordinal or interval data.
Many researchers (but not all) insist that likert scales are ordinal. But I am not one of those researchers. The reason being is that you cannot report means for ordinal data - they are meaningless because there are unequal intervals between the scales. However, with Likert scales, there is an equal distance in the scale (1,2,3,4,5). Which means you can report means and it has meaning. For example if you have a Likert scale for customer satisfaction and where 1=very unsatisfied, 2=somewhat unsatisfied, 3=neither satisfied nor unsatisfied, 4=somewhat satisfied, 5=very satisfied, to have a mean of 3.5 tells us that the majority of people are satisfied. And assuming that the standard deviations were similar, if we were to compare the satisfaction rate between two companies, we could safely say that customers are less satisfied with a company that has a satisfaction rate of 3.4 compared to 3.6. The point being that the mean is useful data.
The reason there is an argument as to whether Likert Scales are ordinal or interval is that researchers say that the labels that respondents choose from (very unsatisfied to very satisfied) does not have a true measurable interval and that we are assuming that all people that select very satisfied are equally distant in their satisfaction from those people who select somewhat satisfied. This is a compelling argument, but in reality, how you view Likert scales depends on your field of research. In my field (Marketing) it is generally accepted that you must report the means of Likert Data (making it an interval scale). This is especially true if you are doing advanced analyses with the scales, such as Structural Equation Modeling, where you will need to report the Likert Scales means in the results section.
My advice to you is to follow the generally accepted stance towards Likert scales for your field and if they follow the stance that Likert data is an ordinal scale, that you can't report the means.
As far as substrating, I am not sure what you mean. Both ordinal and interval data have hierarchies, meaning that there is a natural order where one category is "more" or "less" than the next category. So combining a 3 and a 5 is not recommended. You could possibly take the 5 point scale (1 is strongly disagree, 2 is disagree, 3 is neutral, 4 agree and 5 is strongly agree) and turn it into an ordinal scale by creating a new variable with the categories of disagree=1 (recoding 1s and 2s of original scale into a 1), neutral=2 (recoding the 3's to 2's) and agree=3 (recoding 4s and 5s into 3). This would reduce the scale and make it an ordinal scale, where there is hierarchy, but the distance between the three numbers is most likely uneven. In this case you would not report the mean. There are also other ways to report the Likert data, like using top-box analysis that combines the "satisfied" and reports it as a percentage.
I hope this helps clarify things.
Here are some videos that might be helpful:
ruclips.net/video/XieVy77j12Q/видео.html (Understanding Interval Scales)
ruclips.net/video/nbdRgKLMUEA/видео.html (Creating Summary charts for Likert Data)
ruclips.net/video/fOkpsvSaOGs/видео.html (Creating Top-Box Summaries)
Maam, can you pls upload a video on how to analyse some tests with pure ordinal data.. doing phd in marketing. Regards from INDIA.
Hello! I have some questions about normality test and finding significant difference when your data was from questionnaire with 3 options (a,b, and c). My lecturer asked me to find the significant difference of learning strategies between 2 groups while the data of learning strategies were taken from questionnaire, I'm so confused on how to find it using SPSS. Could you please help me? Also do I need normality test for my data? Thank you
Hi Faradina. I can't tell what type of data you are using by the information you provided, so I can't be very helpful at this time. What are the questions/scales that were used for the variables? Is the questionnaire with 3 options the grouping variable? How was the differences in learning strategies measured?
In regards to the normality tests, it depends on the assignment that was given to you. Sometimes professors give data that has already passed the normality tests. You might want to ask your lecturer to clarify if you need to do a normality test or not.
@@JenniferTaylorPhD the questionnaire is in type of multiple choice items in which each option identifies different strategy (e.g. a: metacognitive, b: cognitive, c: social). I coded the data 1-3 in Excel and spss (values 1 indicating meta, 2 cognitive, and 3 social) and I categorized the data as nominal not ordinal or scales.
The differences were measured by grouping their responses of questionnaire based on their group (two groups in my research are high and low self-efficacy). My lecturer said that I have to find whether the differences of learning strategies between two groups are significant or not yet I'm not familiar with statistical package in SPSS and how to compute it.
Anyway, thanks for replying to my comment.
@@faradinamaple8503 So I think your issue might be one of identifying the predictor and outcome variables, which I think I am still confused by. If I am understanding you correctly, you want to know whether the learning strategies (metacognitive/cognitive/social) have statistically different outcomes. My question is what is your outcome variable? What are you trying to predict? It seems that you are trying to find the statistical differences in Self-efficacy based on the learning strategy. If that is the case, then I am curious as to whether or not you were required to create high/low groups for self-efficacy or if it was something that you thought you needed to do. I say this because you can use SPSS to run an ANOVA using the three nominal groups (metacognitive/cognitive/social) and the dependent (outcome) variable of self-efficacy. Using post hoc tests (such as the Tukey test) you can test the means to see if there are significant differences between the three groups.
@@JenniferTaylorPhD the research question of my research is: is there any significant difference of learning strategies between students with high and low self-efficacy?, I used 5 multiple choice questions with options (a,b,c) to identify their preference of learning strategies. I have two groups (high and low self-efficacy). In SPSS, I categorized all of my data as nominal. However, I'm still confused on how to find the significant difference of learning strategies between two groups. I have done what you suggested (I used one way ANOVA) but the result showed E (errors). I read from other websites, two way ANOVA or chi square test can be used but honestly I still don't know.
@@faradinamaple8503 Thank you for the clarification, this helps me understand the analysis you need to use. Your issue is that your dependent (outcome) variable is categorical (nominal). You can't use ANOVAs with categorical dependent variable because this type of analysis requires that the dependent variable is either interval or ratio (continuous). The reason being is that an ANOVA will test two groups for differences in the mean of the dependent variable. With categorical dependent variables, you would need to use multinomial logistic regression. I think you might want to talk with your Lecturer before you continue, just to make sure that you are on the right path.
You mentioned categorical data being ordinal at 3:09, but isn't categorical data always nominal?
Great question, Eva. At 3:09 I say that they responses are categories. So you have the question and the scale which is organized into categories. The categories may be nominal, ordinal, interval or ratio (but ratio data is typically an open ended question, unless you spend a great deal of type creating all possible categories to choose from).
A question has a nominal scale would have categories that are simply names and have no hierarchy in that one category is not "more" than another categories. For example, "What is your gender?" with a scale of Male/Female/Prefer not to answer - this would have three categories that the respondent could choose from and would be coded in the data as a 1, 2 or 3. You could report percentages of each category, but you could not report mean, median and mode, as there is no hierarchy and therefore the statistics mean nothing.
With ordinal data, you have a question where the scale is organized into hierarchical categories, where one item is "more" than the other, but the amount by which they are "more" is different between each category. For example, "Please select your age range." with a scale of "18-29/30-42/43-60/61+" the categories are arranged from smallest to largest, but the interval between the categories is different (12 year/13 years/18 years/indefinite years) - this makes it an ordinal scale. Therefore you can report on the frequencies of each category, the mode would tell you which category was selected most and the median would give you an idea as to the dispersion, but the mean doesn't really tell us much.
With interval data, you have a question where the scale is organized into hierarchical categories with exactly the same interval between each category. For example, "Which of the following best describes your age range?" with a scale of "0-20/21-40/41-60/61-80/81-100/101-120/120-140" - the trick here is you have to capture all the possible ages and have equal units for each scale, which in this case is 20 units. In this case, you can report the frequencies, mean, standard deviation, median, and mode.
There is debate as to what type of scale a Likert scale is (for example a scale that is "Highly disagree/somewhat disagree/Neutral/somewhat agree/highly agree) an ordinal or interval data. With much of the debate surrounding the ability to report the mean. In my opinion, reporting the mean and standard deviation tells a better story because it tells you about the dispersion of the data. (Where people clustered around one category or across them all). Take two companies that have a satisfaction mean of 4.5 on a 5 point likert scale (1=highly dissatisfied, 2=somewhat dissatisfied, 3=neither, 4=somewhat satisfied, 5=highly satisfied). A 4.5 mean suggests that most people are somewhat or highly satisfied, with a skew towards highly satisfied. But if you add the standard deviation it tells a different story - where 4.5 (SD=.02) says that the data is clustered close together around the somewhat satisfied and highly satisfied. But a mean of 4.5 (SD=1.8), would say that the data is spread across the categories, and therefore there are more dissatisfied people. So who is doing better? The company with the smaller standard deviation. And I can tell this story by simply reporting the statistics and then providing evidence by using the frequency charts as a follow up explanation when reporting to a client.
I hope this explanation helps. Please let me know if you have any more questions.
Good day sir Can I ask a favor from your goodheart😇 may I seek for a help .. on how I am going to put my availbale data if I have the following statement of Problem
1. What is the attitude of respondents toward learning
A. Attitudesowards academics
B. Atiitudes toward teachers and co-pupil
C. Attitudes towards the subject
D. Over-all attitude toward learning
How do Iincode it properly in the SPSS. Sir/madam😇
Thank you for the fast responce Godbless.I am not good at statistic either😅
Hi Ms, Taylor, I am just confused with what type od data are likert scale items in your video. According to what I read, Individual likert scale items are meant to be "ordinal" but in your video it says "scale". May I know which one is more appropriate?
Hi William,
Excellent question. One that actually is hotly debated. And the short answer is that categorizing Likert scales as "ordinal" or "interval" depends on your field.
Likert scales are the trickiest of the scales. They cause the most confusion amongst students and the most infighting between researchers.
Likert scales are scales that use scales with labels that are balanced between positive and negative. Such as scales that measure satisfaction measured as 1=very unsatisfied; 2=somewhat unsatisfied; 3 = neither satisfied nor unsatisfied; 4=somewhat satisfied; 5= very satisfied. Or scales that measure importance, likelihood, agree/disagree.
Many researchers, especially those in the mathematical sciences, insist that likert scales are ordinal. However, in disciples like Marketing we classify them as interval scales. The reason why there is a debate has to deal with reporting means. Remember, with ordinal data you cannot report means. Means are meaningless with ordinal data because there are unequal intervals between the scales. Think about having income brackets that are 1= Under $15k; 2=$15,001 - $50k; 3=$50,001 to $90k; 4= $90 and over. If you report a mean of 2.3, what does it mean? Add in a standard deviation, which measures the spread of the answers, and things get even more meaningless. Thus, we don't report means on ordinal data.
However, social scientists, such as Marketing researchers, argue that Likert scales do indeed have equal distances in the scales, which would make them interval scales. The argument suggests Likert scales have equal distance in the scale (1,2,3,4,5). Which means you can report means and it has meaning. For example if you have a Likert scale for customer satisfaction and where 1=very unsatisfied, 2=somewhat unsatisfied, 3=neither satisfied nor unsatisfied, 4=somewhat satisfied, 5=very satisfied, to have a mean of 3.5 tells us that the majority of people are satisfied. And assuming that the standard deviations were similar, if we were to compare the satisfaction rate between two companies, we could safely say that customers are less satisfied with a company that has a satisfaction rate of 3.4 compared to 3.6. The point being that the mean is useful data.
The reason there is an argument as to whether Likert Scales are ordinal or interval is that researchers say that the labels that respondents choose from (very unsatisfied to very satisfied) does not have a true measurable interval and that we are assuming that all people that select very satisfied are equally distant in their satisfaction from those people who select somewhat satisfied. This is a compelling argument, but in reality, how you view Likert scales depends on your field of research. In Marketing, we assume that all people interpret the scales similarly. This is a terrible assumption for purist mathematician, but for what we do it is generally accepted that Likert scale means have valuable meanings. Therefore, in marketing, you must report the means of Likert Data (making it an interval scale).
If you are not in Marketing, then you need to follow the standard for your field.
I hope this answers your question without causing more confusion. Please let me know if you have any additional questions.
Warm regards,
Jennifer