How to build MaxDiff simulator in standalone application?

Question

Hi Brian

Thank you so much for the promopt reply. I will try this excellent web-based maxdiff analyzer. But I still need your help on developing standalone MaxDiff simulatior as we cannot share this web-based simulator with the client . Therefore, we wish to develop an Excel-based simulator. In this case, would you please kindly introduce some articles of maxdiff algorithm, or let me know if sawtoothsoftware has already developed the standalone maxdiff simulator.

I am highly appreciated your kind support.

Comments

Why can you not share the web-based simulator with your client?  Is it because the client will not let you put the data on the internet?  Or is there some other reason?

---

Hi David, for one reason, it is because we won't share our account with client, or he/she would see all my projects under this account. Do you have any MaxDiff Analyzer for client simulation only (like online conjoint simulator)? For the other reason, it is due to the data sensitivity and security while put the data online. As such, is there any security guarantees that could show to the client?

---

I can create a client account for you if you are interested (email me directly if you would like to know more details about this option - dave@sawtoothsoftware.com).

As for the security guarantees, I can only say that we are following industry standards to keep the information secure.  The Analyzer and the online simulator both use SSL which encrypts the data as it travels across the internet.  We use Rackspace servers and we believe that they are following good security practices.  We feel good about the security of the data.  All of that said, there is no guarantee that it is 100% safe.

---

Hi David, thank you so much for the prompt reply. It is great that you could help to create client account for us. I will drop an email for details once we determine to apply this wonderful online maxdiff analyzer.

Besides, also thankful for your explanation on security aspect. Now, I am confidence to tell my client about the data storage online. I definitely agree that there is no guarantee that everything is 100% safe online.

I will have a trial on MaxDiff Analyzer before we officially apply it in our projects. Thanks.

Answer 1

Send an e-mail into our support at sawtoothsoftware.com address.  I have a sample spreadsheet that shows how to go from raw scores to rescaled scores for respondents.

As far as the simulator part, our MaxDiff Analyzer tool currently does first choice and share of preference simulations.  First choice is just whichever item has the highest score, that's the one you choose.  Share of preference is just like conjoint.  e^u1 / (e^u1 + e^u2) would give you the share of preference for item 1 when compared in a simulation against item 2.

Calculating reach and frequency for TURF is doable as well, you just discretify the data to turn it into 0 and 1, then count it up.  Searching to figure out the best combinations would be a lot more advanced, though I'm sure it would be possible for someone with more knowledge of things like programming and macros in Excel.

Comments

Hi Brian, thanks very much for your feedback. I will send an email to support team then.

Regarding the simulator, we would like to adopt share of preference method. Then, for u1, u2, ..., should I use raw scores or rescaled scores? Meanwhile, could you please explain more about "e^u1 / (e^u1 + e^u2) would give you the share of preference for item 1 when compared in a simulation against item 2"? What do you mean by "compared in a simulation again item 2"? If I have 30 items, should the denominator be "e^u1+e^u2+...+e^u30"?

Regarding TURF calculation, how to discrete the data into 0 and 1? Do you mean if one item is selected, it would be recoded to 1? Or do you have any related documents to share?

I am appreciated your kind feedback on my questions.

---

Raw scores should be used for simulations, while rescaled scores are used when going "across" respondents, such as calculating an average score for item X, or submitting the scores for cluster analysis, etc.

For the share of preference calculations, having all 30 items in the denominator would calculate the probability of choosing that item when compared against all of the other items.  But, if you just wanted to compare choosing item 1 or item 2, then just those items are in the denominator.

For the TURF analysis, you can read a bit about how we "discretify" the item scores in the help file at https://maxdiffanalyzer.com/mda/faces/help/index.xhtml.  I probably have some slides from a workshop I gave last year that has some numeric examples if you wanted more than an explanation.

---

Hi Brian,

I am very appreciated that you replied me promptly. Now, I am much clearer about the usage of raw scores and rescaled scores. Also, it is quite easy understanding the share of preference calculations.

It would be great if you share some slides with me about numeric examples on TURF analysis. I will also write it in email and send to your support team. Furthermore, when I am doing MaxDiff analysis, I found a bit difference for the sequence (from most important to least important) of rescaled items and raw items. I also enclosed the results to my email. Thanks for your time to have a look at it.