Hi David, happy to have found your beginner friendly videos that are easy to understand. Im struggling to know if i run cronbachs alpha on the subscales that were created after i computed all the variables for my questionnaires. I know that if a questionnare measures more than one construct that you individually run cronbachs alpha on them but is this the same for subscales within a questionnaire. Thankyou
Thanks for your question. Only certain items need to be reverse coded. E.g., if you had an extroversion questionnaire and one item was "I love parties" (0 = strongly disagree, 5 = strongly agree") and one item was "I like to keep to myself" (0 = strongly disagree, 5 = strongly agree"), the second item would need to be reverse coded so that higher scores represent higher levels of extroversion.
@@DavidRobinsonPhD Hi, Doc! I have a bit of a confusion, I have non-normally distributed data, specifically positively skewed, and I transformed the data by using log10 to satisfy the assumption. Now, if I'm going to run reliability analysis, should I use the new values (from log10) or the initial values (the actual score of the items)? Thanking in advance!
@@markfrederick6033 Hi Mark, thanks for your question. I think it would be best to perform the reliability analysis with non-transformed data if possible as this is conventional and doing the same may facilitate comparisons with the analyses of others. If you have a large sample, you could support this position by arguing that the test is robust to violations of normality with large samples: www.ncbi.nlm.nih.gov/pmc/articles/PMC3279724/
@@DavidRobinsonPhD Thank you so much for this, and for the included reference. Whilst taking BSc Psychology, I am also in training as a student-researcher, you've helped me plenty of times through your videos. Thanks for the content, Doc! Please, stay safe! Cheers!
Hey mate appreciate your video on this. In an instance a scale made up of 3 sub scales has poor alpha for 2 of the sub scales, but strong total scale alpha - what does this mean? Is the total scale reliability thus misleading - thanks!
Usually, the data associated with Likert scales are treated as ordinal. Here, SPSS automatically labelled the variables as nominal and I didn't relabel them as it doesn't affect the processes or analyses in the video.
Hi. I have an issue regarding the reverse coding. I plan to use it on a continuous variable with a scale from 0-100. I don't want to do this manually. Any chance there is an option of doing it, for instance (1-variable) in spss? Highly appreciate an answer. I'm a bit lost.
Hi, thanks for your question. Unfortunately, I'm not aware of a faster way of doing it. However, I imagine it would only take a few minutes to do it manually.
Thanks, this was very helpful, but I have a question. What happens to the Chronbach alpha value if u don't do a reverse score for the negatively worded statements? does the chronbach alpha improve or reduce?
Hi Timothy, thanks for your question. When high responses on one item are associated with high responses on another item, the alpha is higher. If you had one positively worded item and one negatively worded item and you didn't reverse score the latter, high scores on the former would likely be associated with low scores on the latter. Therefore, failing to reverse score items that should be reverse scored will likely reduce the alpha.
Love the video. I am currently doing my dissertation, I am trying to check the reliability for subscales I have newly created. The items comes from multiple measures - I have recoded them all onto the same Likert scale but keep getting the error message "too few cases". From the help online, I am thinking it is as the missing values are scattered over numerous analysis variables. Id there any way round this? Any advice would be appreciated..........
Sir, Thank you for this informative video. I am working on a questionnaire with 12 sub-scales (each sub-scale with about 2-3 items in it). Will you please tell me a method to work on its reliability? I tried calculating it's Cronbach alpha value but it's coming low. I think, it's because of the number of sub-scales in it.
Hi Amber, this suggests that the items aren't positively correlated. I'd perhaps just double-check that the reverse coding has been completed properly (assuming there are reverse items) and that only items that are supposed to be included in the analysis are included.
@@DavidRobinsonPhD I did recode my items, and I did them properly. Well at least how my professor showed me. I double checked to ensure the items recoded were done right. However, I read that if the sample size is really small it can cause it to come out negative. I have ran these recoded items with my full questionnaire/scale, but once I did my sub scales they came out negative ☹️ it’s for my final so I’d like for it to be right but idk how to fix it.
@@Honeyyy.b_ If you ticked "Scale if item deleted" (I tick this just after 5.00 in the video), you could take a look at the "Cronbach's alpha if item deleted" column in the output to see if any items are problematic and then consider removing them if they are.
Very helpful thanks David, you explained that very clearly
Thanks Rob!
You explained it so well. Thank you
Thanks, glad it helped!
This has helped me a lot! Thank you for your clear and detailed explanation. Suscribed!
Thanks Zurgiff, glad to hear it helped!
thank you so much sir! it helps me a lot 🤍
Glad it helped!
Hi David, happy to have found your beginner friendly videos that are easy to understand. Im struggling to know if i run cronbachs alpha on the subscales that were created after i computed all the variables for my questionnaires. I know that if a questionnare measures more than one construct that you individually run cronbachs alpha on them but is this the same for subscales within a questionnaire. Thankyou
Ok but do we just reverse code certain items or all of the items on a scale? How do we even choose which ones to reverse code?
Thanks for your question. Only certain items need to be reverse coded. E.g., if you had an extroversion questionnaire and one item was "I love parties" (0 = strongly disagree, 5 = strongly agree") and one item was "I like to keep to myself" (0 = strongly disagree, 5 = strongly agree"), the second item would need to be reverse coded so that higher scores represent higher levels of extroversion.
Thank you so much! 💖
Thanks Mark, glad it helped!
@@DavidRobinsonPhD Hi, Doc! I have a bit of a confusion, I have non-normally distributed data, specifically positively skewed, and I transformed the data by using log10 to satisfy the assumption. Now, if I'm going to run reliability analysis, should I use the new values (from log10) or the initial values (the actual score of the items)?
Thanking in advance!
@@markfrederick6033 Hi Mark, thanks for your question. I think it would be best to perform the reliability analysis with non-transformed data if possible as this is conventional and doing the same may facilitate comparisons with the analyses of others. If you have a large sample, you could support this position by arguing that the test is robust to violations of normality with large samples: www.ncbi.nlm.nih.gov/pmc/articles/PMC3279724/
@@DavidRobinsonPhD Thank you so much for this, and for the included reference. Whilst taking BSc Psychology, I am also in training as a student-researcher, you've helped me plenty of times through your videos.
Thanks for the content, Doc! Please, stay safe!
Cheers!
Hey mate appreciate your video on this.
In an instance a scale made up of 3 sub scales has poor alpha for 2 of the sub scales, but strong total scale alpha - what does this mean? Is the total scale reliability thus misleading - thanks!
for overall reliability, stratified alpha is appropriate, instead of alpha chronach.
thanks, Dr. for your help
mu question, here you are using the Norminal not Ordinal for the Likert scale?
Usually, the data associated with Likert scales are treated as ordinal. Here, SPSS automatically labelled the variables as nominal and I didn't relabel them as it doesn't affect the processes or analyses in the video.
Hi.
I have an issue regarding the reverse coding. I plan to use it on a continuous variable with a scale from 0-100. I don't want to do this manually. Any chance there is an option of doing it, for instance (1-variable) in spss?
Highly appreciate an answer. I'm a bit lost.
Hi, thanks for your question. Unfortunately, I'm not aware of a faster way of doing it. However, I imagine it would only take a few minutes to do it manually.
Thanks, this was very helpful, but I have a question. What happens to the Chronbach alpha value if u don't do a reverse score for the negatively worded statements? does the chronbach alpha improve or reduce?
Hi Timothy, thanks for your question. When high responses on one item are associated with high responses on another item, the alpha is higher. If you had one positively worded item and one negatively worded item and you didn't reverse score the latter, high scores on the former would likely be associated with low scores on the latter. Therefore, failing to reverse score items that should be reverse scored will likely reduce the alpha.
Love the video. I am currently doing my dissertation, I am trying to check the reliability for subscales I have newly created. The items comes from multiple measures - I have recoded them all onto the same Likert scale but keep getting the error message "too few cases". From the help online, I am thinking it is as the missing values are scattered over numerous analysis variables. Id there any way round this? Any advice would be appreciated..........
Sir,
Thank you for this informative video.
I am working on a questionnaire with 12 sub-scales (each sub-scale with about 2-3 items in it). Will you please tell me a method to work on its reliability?
I tried calculating it's Cronbach alpha value but it's coming low.
I think, it's because of the number of sub-scales in it.
Sir how to install spss on mac can you explain it
my alpha is negative and idk why
Hi Amber, this suggests that the items aren't positively correlated. I'd perhaps just double-check that the reverse coding has been completed properly (assuming there are reverse items) and that only items that are supposed to be included in the analysis are included.
@@DavidRobinsonPhD I did recode my items, and I did them properly. Well at least how my professor showed me. I double checked to ensure the items recoded were done right. However, I read that if the sample size is really small it can cause it to come out negative. I have ran these recoded items with my full questionnaire/scale, but once I did my sub scales they came out negative ☹️ it’s for my final so I’d like for it to be right but idk how to fix it.
@@Honeyyy.b_ If you ticked "Scale if item deleted" (I tick this just after 5.00 in the video), you could take a look at the "Cronbach's alpha if item deleted" column in the output to see if any items are problematic and then consider removing them if they are.
Thanks for this informative video