Discrete Choice Experiment (Discrete Choice Conjoint) - A fast analysis
ฝัง
- เผยแพร่เมื่อ 14 พ.ย. 2021
- In this video, I advise my students on how to analyse a DCE.
If you are interested in videos on DCE development, write in the comments below. With a high number of requests, I will make a video.
Otherwise, feel free to listen to my podcasts on DCE:
What is a DCE: mci4research.podbean.com/e/re...
How to develop levels for a DCE: mci4research.podbean.com/e/re...
How to deal with abstract constructs: mci4research.podbean.com/e/re...
What software to use:
mci4research.podbean.com/e/re...
Online tools for a DCE:
mci4research.podbean.com/e/re...
Analysing a DCE:
mci4research.podbean.com/e/re...
Thank you very much for the video and the extra support. It is so much appreciated!!
Welcome! Good luck with your study!
Thank you for sharing this tutorial! Made our second analysis much more easier!
Great video, thank you for the insights !
Dear Eugen, thank you for this amazing video! It was really helpful for us to get an overview on how to analyze our discrete choice analysis. We are left with one question. Maybe you can help us: How does the analysis procedure differ, if we showed three final products, instead of two, to the participants (so the participants had to decide to pick one of three different products with different attributes several times).
Maybe you have some idea or inspiration for us or even have experience with this structure. Have a nice week and best wishes.
Soraya
The difference is minimal - you will have to calculate the final_decision differently as you have the values 0, 1, 2 and not only 0 and 1. And restructuring second part- you will have 3 options seen, so put three variables in one category. Otherwise, all the same!
Sorry, both parts of restructuring need to be adapted. But the logic is the same.
@@eugenleo Thank you so much for your quick answer! This is really helpful. :)
Am I right that we would need to do a multinomial logistic regression instead of a binary logistic regression in this case, because we have three possible values for final_decision?
No, the variable final_decision will be binary (1 - the want the option, 0 - they don't want it). So, it will be a binary logistic regression, I think.
Thank you Eugen! 👍
I used this video to clean and transform my research output, and it worked. Thank you! (I even figured out a problem I had because three of my attributes had the same names - "low", "medium", and "high". the solution is to add an "and" to the if function and connect attribute name to choice)
Indeed, I had this issue recently too. The check variable helps indicate the issue. Thank you and good luck with your research!
@@eugenleo Do you have a video on this? I use yes and no for different attributes. How do I solve this problem? I do not understand what is meant by 'add an and to the if function'.
Thank you in advance!
@@lelavanosselaer8884 Not at the moment. But here is the code that we used:
IF (attr1 = "Competitiveness" & attr1fop = "high" | attr2 = "Competitiveness" & attr2fop = "high" | attr3 =
"Competitiveness" & attr3fop = "high" | attr4 = "Competitiveness" & attr4fop = "high" | attr5 = "Competitiveness" & attr5fop
= "high" | attr6 = "Competitiveness" & attr6fop = "high" | attr7 = "Competitiveness" & attr7fop = "high" | attr8
= "Competitiveness" & attr8fop = "high" | attr9 = "Competitiveness" & attr9fop = "high" | attr10 = "Competitiveness" &
attr10fop = "high") Competitiveness_high=1.
We had the same anserws for the variables "Competitiveness", "Ambition", etc. These variables had the same values - no, low, medium, and high. So we needed to go through each of them and code accordingly.
I hope it helps! I'll try to record a video on this.
@@lelavanosselaer8884 Yes (it is still uploading, but soon you will be able to watch it): th-cam.com/video/L9yKY11A-Tc/w-d-xo.html
@@eugenleo Dear Sir Bogodistov, thank you very much for this video!
Hello, thanks for such a detailed video. I have a question do you know what to when we use choice cards as a picture including atributes and levels. and in the qualtrics, the data that we have is only option A, B and neither. so how can I link the attributes to these options? thanks
Well... This is now a lot of manual work. Practically, you need to create a set of variables for each level within each attribute. It is what I show in the second part of the video. Then, you manually code them for each participant - has s/he seen the level - 1, if not - 0. These dummies will be your independent variables, and the decision - dependent. In the same way you code your dependent variable - if decision option selected - 1, if not - 0. I did it in this way once. It took a while but it is possible. For insurance, if all participants saw the same card - you can code dummies for all participants for this card. Good luck and be patient;)
Dear Bogodistov, thank you for the videos. But I have some confusion about some terminologies. Louviere et al. (2010) have article "Discrete choice experiments are not conjoint analysis. But as you mentioned, some of the articles use discrete choice experiments synonyms with discrete choice conjoint. Can you give me some references to understand this situation or can you explain to me? Thanks a lot.
It is a good question, thank you. As far as I understand, DCE is the art of conjoint. The selection options are presented in a different way - instead of ranking the item, you present the in a form of a manipulation - the participant sees only the two (or more) products from which she has to select one. This different way of presentation makes DCE to an experiment which may be seen as a different method. But (to the best of my understanding) the logic behind the method is pretty the same - to find out the preferences of the participant.
@@eugenleo Thank you so much.
interesting... please add tutorial files in the description as well..
I would like to do it, but the data comes from a real study which has not been published yet. Maybe we will add the data after it will have been published.
would really help to have to files
@@wildgameestate425 I plan to upload the files as soon as the project is published. But it might take a while...
@@eugenleo Any chance you could upload the data?
Thank you for the helpful video. Can you show me how to calculate the marginal effects in SPSS?
Actually, SPSS is not good for this. We applied STATA. But you might be interested in this video which explains how to convert odds ratio to probability (th-cam.com/video/UaCB3AWOWtU/w-d-xo.html) or this video which explains how to convert odds ratio to confidence intervals (th-cam.com/video/G5sqDwiwm68/w-d-xo.htmlsi=W5c64U-96KWvChNa). Otherwise, you could run the data after restructuring in STATA (www.peretaberner.eu/how-to-estimate-and-interpret-marginal-effects-from-logit-model-with-stata/). Good luck!
Thank you so much@@eugenleo
As your question was good and indicated demand for marginal effects calculation, I developed a video for this. Enjoy: th-cam.com/video/5YqAgIiHRCM/w-d-xo.html
May I know how to create decession variable?
I assume tha you mean "decision" variable...
You can find it in 13'30'' of the video. In Qualtrics, you indicate your preference. This preference is saved as the answer to the question as 1 (Option 1) or 2 (Option 2). You use these variables as the dependent variable. In the first restructuring you create the variable "decision" were you put the answers from Qualtrics.
Yet, after the second restructuring, you have to convert these values into 1 or 0. If the Option 1 was selected, the according option should become 1 and 0 otherwise. I do it in 45'25'' of the video.
I hope it helps you with the analysis! Good luck!