Before you run a Cronbach’s alpha or factor analysis on scale items, it’s generally a good idea to reverse code items that are negatively worded so that a high value indicates the same type of response on every item.
So for example let’s say you have 20 items each on a 1 to 7 scale. For most items, a 7 may indicate a positive attitude toward some issue, but for a few items, a 1 indicates a positive attitude.
I want to show you a very quick and easy way to reverse code them using a single command line. This works in any software.
Rather than specifying each individual recoded value–a 1 to 7, 2 to 6, and so on, just subtract the values from a constant one value higher than the highest value on the scale.
For example if OldVariable is reverse coded and on a 1 to 7 scale, in SPSS, do this:
COMPUTE NewVariable = 8 – OldVariable. (You can also do it in the menus in Transform–>Compute).
In SAS, do this within a data step.
Data scale;
Set scale;
NewVariable = 8 – OldVariable;
Run;
The value from which you subtract your old variable will always be one value higher than the highest value you have. So I subtracted my old variable from 8 because I have a 1 to 7 scale. If I had a 1 to 5 scale, I would subtract my old variable from 6.
You can see how it works:
8-7=1
8-6=2
8-5=3
8-4=4
8-3=5
8-2=6
8-1=7
If you only have to reverse code one item, this isn’t a big deal. But I have found that data cleaning and creating new variables often is the step in data analysis that takes the longest. I use whatever shortcuts I can.
Hey, can you please tell me the steps on SPSS? Thank you in advance
Hi,
The data that i’m dealing with is measured on a scale of 0 to 4. For some variables, 0 means bad and 4 means good, and for others, 4 means bad and 0 means good. Can I still use the same subtraction method that you’re describing if my measure includes 0? Maybe I should first 1 add everything so it’s a 1 to 5 scale, and then follow your steps?
Also, even though the methodology describes the data only being collected from 0 to 4, the max observation is 4.013. So if I do the subtraction, should it be from 5 or from 5.013?
Just for total clarity, the scale the data set uses is 0, 1, 2, 3, 4
Thank you for this super helpful blog post!
Hi Sam,
You can use the same method.
And I’m not sure how you can get a max observation of 4.013. I would check that for an error.
Very useful, I needed to do this to some NDVI values but I neither knew how to do nor even know the name it’s called. Some random google search landed me here and it’s a solution to my quest.
Thanks so much.
Just a quick thanks from a survey researcher who was given a hard copy table set and that’s all. I was not a part of the design, and they had a ranked mean reversed. When will people learn that it takes a professional to design a survey!?!
Duh….lol…I was subtracting from 7. Thank you Dr. Karen. Once I got help from you I realized who you were by seeing the book add under “Read Our Book”…I have it right next to me as I am writing this.
Great book when I was a student.
Alpha still negative !!!!/ even after reversing the negative wording, the sample is 900, the question is multiple responses (includes 5 sub-questions) wither it is selected or not. what to do? please ?
How to reverse coding continuous variables?
Thanks
My questions in negative form like how often you faced problem from dampness? Scale i used 1 never and 6 very often.. So most of answer came near 2 or 3… Then can i use reverse coding?
If appears a negative factor loading in second order exploratory factor analysis, what I do? Can I reverse de coding of that first order factor? If yes, how I do that?
My main goal is to calculate the internal consistency of a second order factor, but on of the loadings gave a negative value…
how I categorical in 5 likert scale in this frequencies. 1 strongly agree
2 Agree 3 Neither agree or disagree 4 Disagree 5 Strongly disagree
Valid 2.00
3.00
4.00
5.00
Total 120
I’m doing a pre-post test for students in a class using different methods for half the class sessions – same students. I want to compare which collection methods are received or appreciated the most. How can I use a paired t-test for this comparison?
MAL
Thanks! That saved me a lot of time.
Hello.. can you teach me how to compute data which got question in Likert scale? how to compute the score to the answer as 1 = "Strongly disagree" , 2 = "Disagree", 3 = "Agree", 4 = "Strongly Agree"Thank You.
The way you propose is the quickest ro reverse the direction of an item. I use it often.
If you have missing values you have to complete the action with a new assignment for missing values.
Good point. You’re creating a new variable so it does need new missing value assignments.
Hello,
Agreed data cleaning is very challenging. So my colleague and I have followed the steps to reverse code into a different and new variable, then when we computed the recoded variables and other unchanged variables our output seemed wrong. Are there any additional steps we may have missed, I have made efforts to do research, but I am wondering if I am not using the correct language when research. Any guidance would help tremendously.
Thank you,
Cynthia
Did you use brackets if you were doing a bunch of items at once? Man, it pains me to go back…
Cool! Thank you very much!
Hello –
How do I label items on a sub-scale that were first (1) reversed scored by recoding into a different variable, (2) then were averaged (Transform > Compute, etc)? Do I use the label from the reversed scored items, or the labels from the not-reversed scored items? For example, of my 10 items on the IPIP 50-item scale, for “Extroversion” 5 items were reversed scored. When I get the MEAN of those items, in the Variable view, under “Label”, how do I label that new scale on a scale of 1 – 5? Thanks so much for your help!
Ruth
this might work, if your intention is: high number on a scale is a good thing. but if your scale is supposed to measure sexist attitudes, you better inverse the positive worded items.
this equals a low mean value on the whole scale, indicating that those low on the scale are not sexist, those high are!
Well, yes, that’s the whole point of recoding the direction of the scale.
What if my categories have names like “agree” “somewhat agree” “somewhat disagree” and “disagree”? How do I keep the names without transforming the categories into number?
Hi Lise,
You can still recode them, but you’ll have to do it the long way. Recode each old value into a new value.
Thank you sooo much! So simple.
This saved me a bunch of time! Thanks!
Awesome. 🙂
So clever. Thanks!
The method you have presented works well when the lowest possible scale value is 1. The more general method for reverse scoring would be:
reversed score = (minimum score) + (maximum score) – actual score
Ah, yes! Excellent.
Karen