How do you do regression using continuous data as dependent variable and nonparametric variables (i.e., ordinal data) as independent variables? For instance, I have maximum price buyers are willing to pay for a product as the dependent variable (Y) and the factors considered [ordinal data; ranked from very important (1) to not important (4)] as the independent variables (X1, X2, X3...). Wondering how I can do analyses on such data. Thank you in advance!
02:24, Q6_1: "To what extent do you agree or disagree with the following statements" (1=somewhat disagree, 5 = strongly agree). Can you please explain why this would be considered as interval data? For me it looks like ordinal data because the distances between the categories cannot be considered to be the same. Thank you in advance!
Many researchers (but not all) insist that likert scales, like the one you mention, 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. Your awareness of the distance between categories is the exact argument that divides researchers. Some researchers say that the labels that respondents choose from (Strongly agree to strongly disagree) 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, then you can't report the means.
@@JenniferTaylorPhD Thank you very much for your fast response to my answer. Still, one question is open to me: If, for example, I have a questionnaire and I am using a Likert scale (let's say from 1=very unsatisfied to 5 = very satisfied) and these data are then viewed as ordinal-scaled in my research field, unfortunately I can never use sum values or mean values for further calculations (e.g. regressions, ANOVAS etc.). That means, if I see a Likert scale as ordinal, almost any further calculation (apart from non-parametric methods) is restricted, right? Because with ordinal data I can't do an ANOVA for example. So... what you mean: if possible, it's better to deal likert-scales as metric/interval scaled right? Thank you in advance!
@@manuelleitner1996 While you can run t-tests or ANOVAS on ordinal data, it is questionable whether or not this is appropriate because there is no information on the distance between points on the scale. In my field, we make the assumption that likert scale data is equidistant, and thus it is interval scales. Which means we can report the mean and use the measurement items in advanced analysis (Hayes process, SEM, etc.). In other fields, they do not. In these fields, they analyze ordinal data with nonparametric analyses.
@@JenniferTaylorPhD Okay, thank you very much for your response, it was very nice to talk to you about that interesting topic. Thank you for your help.
The last two variables you mentioned at the beginning of the video (3:30minutes) you discuss (income and volunteer for organization) they are both uneven but you claim one is interval and the other one is ordinal. How can that be possible????? Could you explain? if they are both uneven they should be both ordinals, right???? Do you have an explanation for that?
Thank you for taking the time to comment. I can't find where I state that income/volunteer for organization is interval. But you are exactly right. If they are unequal groups, with a hierarchy, then it would be an ordinal variable.
I have a question. You said that if we have categories of income that are not equal (e.g. 1-15000, 16-50000) then it is considered ordinal. Does it mean that if the categories were equal (e.g. 1-10000, 10000-20000 etc.) then it would not be ordinal? I am little confused because that would still be categories ordered in some kind of rank. Can you please explain that?
Ordinal and Interval scales both have a hierarchy. The difference is that interval scales have identical units within the scale (i.e, a scale for the question, how many hours a day do you sleep? For an interval scale it would be 1-4 hours, 5-8 hours, 9-12 hours, 13-16 hours, 17-20 hours, and 21-24 hours, where the intervals are 4 hours. Or you could have an ordinal scale of 1-4 hours, 5-9 hours, 10-18 hours, 19-24 hours. There is still a hierachy in the ordinal scale, where 5-9 hours is larger than 1-4 hours, but the intervals are 4 hours, 5 hours, 9 hours and 6 hours - which is not an equal interval. It might be helpful to watch these videos to help you understand the difference between scales: Understanding measurement scales workshop: ruclips.net/video/wFtasSX6G4M/видео.html Ordinal Scales: ruclips.net/video/wFtasSX6G4M/видео.html Interval Scales: ruclips.net/video/XieVy77j12Q/видео.html
@@JenniferTaylorPhD Thank you very much for the answer. I found some examples where the variables were considered still ordinal regardless to the fact that the intervals are equal. For example, age divided into categories: 10-20,20-30,30-40, and they said it is ordinal. Is it a mistake then? Or just something there are different opinions about?
What are you trying to do? Are you doing descriptive research, where you just need to report the reposnes or are you doing advance causal research where you want to show causation between variables? This video is for descriptive analysis.
If you have two variable that are at least ordinal, then yes. Correlation matrices are used to determine the strength and direction of relationships between variables. Both variables must pass these two assumptions before you can conduct a correlation matrices: 1) they are at least ordinal--they may also be interval or ratio, but NOT nominal; and 2) they must be monotonic, meaning that they move together. Here is a link that explains it in more detail --- statistics.laerd.com/spss-tutorials/spearmans-rank-order-correlation-using-spss-statistics.php
4:39 "median and mode didn't mean much"
this is accidentally a great pun
How do you do regression using continuous data as dependent variable and nonparametric variables (i.e., ordinal data) as independent variables? For instance, I have maximum price buyers are willing to pay for a product as the dependent variable (Y) and the factors considered [ordinal data; ranked from very important (1) to not important (4)] as the independent variables (X1, X2, X3...). Wondering how I can do analyses on such data. Thank you in advance!
02:24, Q6_1: "To what extent do you agree or disagree with the following statements" (1=somewhat disagree, 5 = strongly agree). Can you please explain why this would be considered as interval data? For me it looks like ordinal data because the distances between the categories cannot be considered to be the same.
Thank you in advance!
Many researchers (but not all) insist that likert scales, like the one you mention, 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.
Your awareness of the distance between categories is the exact argument that divides researchers. Some researchers say that the labels that respondents choose from (Strongly agree to strongly disagree) 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, then you can't report the means.
@@JenniferTaylorPhD Thank you very much for your fast response to my answer. Still, one question is open to me:
If, for example, I have a questionnaire and I am using a Likert scale (let's say from 1=very unsatisfied to 5 = very satisfied) and these data are then viewed as ordinal-scaled in my research field, unfortunately I can never use sum values or mean values for further calculations (e.g. regressions, ANOVAS etc.). That means, if I see a Likert scale as ordinal, almost any further calculation (apart from non-parametric methods) is restricted, right? Because with ordinal data I can't do an ANOVA for example.
So... what you mean: if possible, it's better to deal likert-scales as metric/interval scaled right?
Thank you in advance!
@@manuelleitner1996 While you can run t-tests or ANOVAS on ordinal data, it is questionable whether or not this is appropriate because there is no information on the distance between points on the scale. In my field, we make the assumption that likert scale data is equidistant, and thus it is interval scales. Which means we can report the mean and use the measurement items in advanced analysis (Hayes process, SEM, etc.). In other fields, they do not. In these fields, they analyze ordinal data with nonparametric analyses.
@@JenniferTaylorPhD Okay, thank you very much for your response, it was very nice to talk to you about that interesting topic. Thank you for your help.
@@JenniferTaylorPhD awesome
This was very helpful thanks 🙏
The last two variables you mentioned at the beginning of the video (3:30minutes) you discuss (income and volunteer for organization) they are both uneven but you claim one is interval and the other one is ordinal. How can that be possible????? Could you explain? if they are both uneven they should be both ordinals, right???? Do you have an explanation for that?
Thank you for taking the time to comment. I can't find where I state that income/volunteer for organization is interval. But you are exactly right. If they are unequal groups, with a hierarchy, then it would be an ordinal variable.
Thank you very much! 💖
you area awesome. thanks
I have a question. You said that if we have categories of income that are not equal (e.g. 1-15000, 16-50000) then it is considered ordinal. Does it mean that if the categories were equal (e.g. 1-10000, 10000-20000 etc.) then it would not be ordinal? I am little confused because that would still be categories ordered in some kind of rank. Can you please explain that?
Ordinal and Interval scales both have a hierarchy. The difference is that interval scales have identical units within the scale (i.e, a scale for the question, how many hours a day do you sleep? For an interval scale it would be 1-4 hours, 5-8 hours, 9-12 hours, 13-16 hours, 17-20 hours, and 21-24 hours, where the intervals are 4 hours. Or you could have an ordinal scale of 1-4 hours, 5-9 hours, 10-18 hours, 19-24 hours. There is still a hierachy in the ordinal scale, where 5-9 hours is larger than 1-4 hours, but the intervals are 4 hours, 5 hours, 9 hours and 6 hours - which is not an equal interval.
It might be helpful to watch these videos to help you understand the difference between scales:
Understanding measurement scales workshop: ruclips.net/video/wFtasSX6G4M/видео.html
Ordinal Scales: ruclips.net/video/wFtasSX6G4M/видео.html
Interval Scales: ruclips.net/video/XieVy77j12Q/видео.html
@@JenniferTaylorPhD Thank you very much for the answer. I found some examples where the variables were considered still ordinal regardless to the fact that the intervals are equal. For example, age divided into categories: 10-20,20-30,30-40, and they said it is ordinal. Is it a mistake then? Or just something there are different opinions about?
I have two ordinal variables. I am using a questionnaire and medications. I do no know what to do. 😭
What are you trying to do? Are you doing descriptive research, where you just need to report the reposnes or are you doing advance causal research where you want to show causation between variables? This video is for descriptive analysis.
Thank youuu
Can we use correlation and covariance matrix for ordinal data ?
If you have two variable that are at least ordinal, then yes. Correlation matrices are used to determine the strength and direction of relationships between variables. Both variables must pass these two assumptions before you can conduct a correlation matrices: 1) they are at least ordinal--they may also be interval or ratio, but NOT nominal; and 2) they must be monotonic, meaning that they move together. Here is a link that explains it in more detail --- statistics.laerd.com/spss-tutorials/spearmans-rank-order-correlation-using-spss-statistics.php
Why didn't you calculate mean in this case?
@@priyankamainali7425 Because it is an ordinal variable which mean has no use for interpretation, median is prefered instead.