if we had 1000 respondents, do we need to input each respondents' data manually at 20:00??? isnt there a way to import the data from Qualtrics? or automate the process?
Very nicely done - I like your straight-forward style. One question. At the very end you indicate what each of the coefficients indicate in terms of what 3333's preferences were. This is easy with the discrete variables, because you're just picking the highest coefficient. But, for the battery variable, which is modeled as continuous, you indicated that you thought the -0.86 coefficient indicated preference for a battery life of 6 months. At what point would you say that the user's preference would indicate 12 months, and 16 months. Obviously the coefficient would need to be positive for you to label 3333's preference as 18 months, but what would the coefficient's value have to be for that to be the case? And same question for 12 months? thanks!
Hi Mark, in this example we made the analytical decision to model battery life as linear. Thus the parameter (coefficient) of -0.86 would be interpreted literally the same way you would a linear predictor in, say, a regression model (as BAT LIFE increases by 1 unit, utility decreases by -0.86 units). If we further assume that 6 months to 18 months is the only feasible range for the battery life (managerial/engineering constraint), we thus simply assume that the lowest valid input value (since the coefficient was negative -0.86) is "most" desired by the individual. The bar charts here are actually not a great visualization... ideally SPSS would depict this result as a line chart (since that is more consistent with our choice to model this attribute as having a linear influence on utility). NOTE: In this illustration I made battery life linear for the sake of more broadly illustrating how to interpret the results (pedagogical consideration). In the REAL WORLD, it is my experience that *all* inputs (even those that could be treated as linear or curvilinear) are just assumed to be categorical (discrete) inputs. The upside of this REAL WORLD application is there is no math form imposed onto the results, the downside being you can't model "in between" values from the results (like 7, 8, 9 or 10 battery life hours, for example).
Thank you for the informational video. I have a doubt while running the syntax file, it is showing an error that one of holdout card contains a value that is not present in the original card. Plz, can you help sir?
In very big strokes, yes, but not exactly. For choice-based conjoint, the underlying analytical model is going to be something for a different type of dependent variable (some sort of model for discrete choice), thus the application is going to be a bit different. And, since a lot of choice-based conjoint models today have some nice features that allow for adaptive, in process, updating of attribute parameters as a person proceeds through the conjoint task, the underlying math model is *much* more sophisticated than what I show here. Check out Sawtooth Software and some of their papers describing this.
Hi Dr. Baker! Instead of using a price or a rating, what if I got only the sequence/ranking of products? would it be exactly the same thing? For instance, let's say the interviewed one needed to choose products in a supermarket shelf, and he first chose the product 'A', then I removed all the products A from the shelf and now he chooses the product 'C'. How should I input this data? Also, is the SPSS able to run the example that I provided, but with hundreds of people been interviewed? tks a lot
In terms of implementing the conjoint experiment, the idea is much the same... the difference here is in your dependent variable (I purposefully used a more straightforward continuous variable in my DV to make the illustration). It would NOT be technically right to just treat the rankings as actual continuous values, but I also wouldn't be shocked if you get mostly similar findings and insights if you treat it that way. More appropriately, you would want to model your DV using some sort of regression option that is for ORDINAL data (a little googling will show you that there are a few options for regression models when your DV is ordinal). Without getting deeper into the weeds, that's my initial thoughts! Good luck! As for SPSS... I don't think there is any reason SPSS can't handle it!
Hi Dr Baker, thank you for this video. I am facing a different research problem. I have 9 types of services offered to customers (in 9 columns). In the 10th column I have the profitability of each client to the company calculated in US dollard. I have 400 rows akin to the 400 customers. I would like to use conjoint analysis to assess the weight or utility of each services so that the company will reward the customers according to the weight of each service. In other words, I would like to know what is the coefficient of importance of each service so that the company will reward each customer based on the weight of each service. The weight being measured by the profitability. I look forward to hearing back from you.
Dr. Baker, thanks so much for your videos!! they are awsome! and this one particularly helped me to go through the Conjoint analysis! I saw at least 6 tutorials from other authors and yours is the one that saved my results! thanks thanks thanks!!!
Hi Dear Dr.Baker Thank you for such a good video on conjoint analysis. After running the syntax I got the error about UTILITY command as "Text: utility.sav Command: CONJOINT The filename is not valid. Execution of this command stops." I Would appreciate if you help me to find the problem?
I'm actually curious as to how SPSS figured out the 9 combinations that needed to be used for the experiment, out of the 18. Can someone explain this me?
I'm not the "math guy" who can actually do the proof, but the logic is "every level of each attribute must interact with each level of each other attribute at least once, try to minimize the combinations under this constraint"... there's actually multiple solutions that satisfy this requirement, the one in this particular video just happens to be the one that iteration found (and by setting the same seed value as I did, you would also find this particular solution).
Manually editing the Syntax file generated is entirely possible! The syntax file you see arounf minute 8:30, for example, could be entirely written by hand or edited as needed.
Good explanation of step by step for Conjoint Analysis. Thank you for your efforts, Prof!
if we had 1000 respondents, do we need to input each respondents' data manually at 20:00??? isnt there a way to import the data from Qualtrics? or automate the process?
Very nicely done - I like your straight-forward style. One question. At the very end you indicate what each of the coefficients indicate in terms of what 3333's preferences were. This is easy with the discrete variables, because you're just picking the highest coefficient. But, for the battery variable, which is modeled as continuous, you indicated that you thought the -0.86 coefficient indicated preference for a battery life of 6 months. At what point would you say that the user's preference would indicate 12 months, and 16 months. Obviously the coefficient would need to be positive for you to label 3333's preference as 18 months, but what would the coefficient's value have to be for that to be the case? And same question for 12 months? thanks!
Hi Mark, in this example we made the analytical decision to model battery life as linear. Thus the parameter (coefficient) of -0.86 would be interpreted literally the same way you would a linear predictor in, say, a regression model (as BAT LIFE increases by 1 unit, utility decreases by -0.86 units). If we further assume that 6 months to 18 months is the only feasible range for the battery life (managerial/engineering constraint), we thus simply assume that the lowest valid input value (since the coefficient was negative -0.86) is "most" desired by the individual. The bar charts here are actually not a great visualization... ideally SPSS would depict this result as a line chart (since that is more consistent with our choice to model this attribute as having a linear influence on utility).
NOTE: In this illustration I made battery life linear for the sake of more broadly illustrating how to interpret the results (pedagogical consideration). In the REAL WORLD, it is my experience that *all* inputs (even those that could be treated as linear or curvilinear) are just assumed to be categorical (discrete) inputs. The upside of this REAL WORLD application is there is no math form imposed onto the results, the downside being you can't model "in between" values from the results (like 7, 8, 9 or 10 battery life hours, for example).
Thank you for the informational video. I have a doubt while running the syntax file, it is showing an error that one of holdout card contains a value that is not present in the original card. Plz, can you help sir?
Thanks for the video, can I use the same method for the choice based conjoint analysis?
In very big strokes, yes, but not exactly. For choice-based conjoint, the underlying analytical model is going to be something for a different type of dependent variable (some sort of model for discrete choice), thus the application is going to be a bit different. And, since a lot of choice-based conjoint models today have some nice features that allow for adaptive, in process, updating of attribute parameters as a person proceeds through the conjoint task, the underlying math model is *much* more sophisticated than what I show here. Check out Sawtooth Software and some of their papers describing this.
Dear Sir,
What are rules which spss follow to generate number of cases i.e why number of cases generated are 9, why not 8 or 10
Hi Dr. Baker! Instead of using a price or a rating, what if I got only the sequence/ranking of products? would it be exactly the same thing? For instance, let's say the interviewed one needed to choose products in a supermarket shelf, and he first chose the product 'A', then I removed all the products A from the shelf and now he chooses the product 'C'. How should I input this data? Also, is the SPSS able to run the example that I provided, but with hundreds of people been interviewed? tks a lot
In terms of implementing the conjoint experiment, the idea is much the same... the difference here is in your dependent variable (I purposefully used a more straightforward continuous variable in my DV to make the illustration). It would NOT be technically right to just treat the rankings as actual continuous values, but I also wouldn't be shocked if you get mostly similar findings and insights if you treat it that way. More appropriately, you would want to model your DV using some sort of regression option that is for ORDINAL data (a little googling will show you that there are a few options for regression models when your DV is ordinal). Without getting deeper into the weeds, that's my initial thoughts! Good luck! As for SPSS... I don't think there is any reason SPSS can't handle it!
@@DrBakerSDSU thanks for sharing your thoughts! You helped me a lot! Best regards!
Very helpful! only one question, where can I find the syntax documents? in order to just copy and paste into the spss editor?
in spss sample file: there a file name conjoint.sps - anyway great video Dr Baker. Thanks
Hi Dr Baker, thank you for this video.
I am facing a different research problem. I have 9 types of services offered to customers (in 9 columns). In the 10th column I have the profitability of each client to the company calculated in US dollard. I have 400 rows akin to the 400 customers. I would like to use conjoint analysis to assess the weight or utility of each services so that the company will reward the customers according to the weight of each service. In other words, I would like to know what is the coefficient of importance of each service so that the company will reward each customer based on the weight of each service. The weight being measured by the profitability.
I look forward to hearing back from you.
Dr. Baker, thanks so much for your videos!! they are awsome! and this one particularly helped me to go through the Conjoint analysis! I saw at least 6 tutorials from other authors and yours is the one that saved my results! thanks thanks thanks!!!
Glad this was of some use to you!
what if we use rank instead of rate/score?
Hi Dear Dr.Baker
Thank you for such a good video on conjoint analysis. After running the syntax I got the error about UTILITY command as
"Text: utility.sav Command: CONJOINT
The filename is not valid.
Execution of this command stops."
I Would appreciate if you help me to find the problem?
Ehsan, hard to say without more context. Is it possible you're using a lighter version of SPSs that doesn't include the CONJOINT feature?
I'm actually curious as to how SPSS figured out the 9 combinations that needed to be used for the experiment, out of the 18. Can someone explain this me?
I'm not the "math guy" who can actually do the proof, but the logic is "every level of each attribute must interact with each level of each other attribute at least once, try to minimize the combinations under this constraint"... there's actually multiple solutions that satisfy this requirement, the one in this particular video just happens to be the one that iteration found (and by setting the same seed value as I did, you would also find this particular solution).
hi sir what is seed value?
Thanks for the video. How can we edit the level of attributes later on?
Manually editing the Syntax file generated is entirely possible! The syntax file you see arounf minute 8:30, for example, could be entirely written by hand or edited as needed.
Nice explanation
16.43 How to generate those cards on Qualtrics?