Survey Design Help

#1
Hi Everyone,



I am having a lot of trouble figuring out the runs for a 2*3 4*4 survey design. From the following site, it is orthogonal at 16 runs, so I don't think I am trying to do something impossible.

http://neilsloane.com/doc/cent4.html



But for the life of me I can't figure out what the runs are. I've tried a bunch of different iterations but resultant regressions won't run properly because the parameters aren't set up properly. Any idea what I am doing wrong?



Below Brand/Price/HP and MPG have 4 possible values. Var1, Var2, and Var have 2.


Run Brand Price HP MPG Var1 Var2 Var3
1 Brand1 Price1 HP1 MPG1 Var1Off Var2Off Var3Off
2 Brand1 Price2 HP2 MPG2 Var1On Var2On Var3On
3 Brand1 Price3 HP3 MPG3 Var1Off Var2On Var3On
4 Brand1 Price4 HP4 MPG4 Var1On Var2Off Var3On
5 Brand2 Price1 HP2 MPG3 Var1On Var2On Var3Off
6 Brand2 Price2 HP1 MPG4 Var1Off Var2On Var3Off
7 Brand2 Price3 HP4 MPG1 Var1On Var2Off Var3On
8 Brand2 Price4 HP3 MPG2 Var1Off Var2Off Var3Off
9 Brand3 Price1 HP3 MPG4 Var1Off Var2On Var3On
10 Brand3 Price2 HP4 MPG3 Var1On Var2Off Var3Off
11 Brand3 Price3 HP1 MPG2 Var1Off Var2Off Var3Off
12 Brand3 Price4 HP2 MPG1 Var1On Var2On Var3Off
13 Brand4 Price1 HP4 MPG2 Var1On Var2Off Var3On
14 Brand4 Price2 HP3 MPG1 Var1Off Var2Off Var3On
15 Brand4 Price3 HP2 MPG4 Var1On Var2On Var3Off
16 Brand4 Price4 HP1 MPG3 Var1Off Var2On Var3On


I would greatly appreciate any help, I am going crazy trying to figure this out!
 
#2
Are you trying to set up a ratings-based conjoint analysis, and these are the 16 cards being rated?

How do you have the IVs set up in the data file?
 
Last edited:
#3
Hi Injektilo,

Yes, respondent rates the concepts from 1 - 5

So formula is:
Rating = Brand + Price + HP + MPG + Var1 + Var2 + Var3

I have the IV set up as 1,2,3, etc

So a row inputted into R might be
5 1 2 3 0 1 0 0

So person rated the concept a 5, which included the first brand, second price, 3rd HP, 4th MPG, Var 1 is yes, Var 2 is no, Var 3 is no

That help?
 
#4
The problem is that you're treating the IVs as continuous variables, when in fact they're categorical. So you first need to dummy code your IVs, so that each level is its own IV and it's either a 0 if it's not present, or 1 if it's present. Then you have to remove one of the levels (typically the first one) to avoid linear dependency.

I've attached an Excel file that you can use as an example (I hope this works, I've never tried it before). In this example, respondents rate 9 cards in a conjoint exercise with 3 attributes and 3 levels each: Brand (VISA, MasterCard, American Express), Interest rate (10%, 15%, 20%), and Credit limit ($2,500, $5,000, $10,000). Notice that the first level of each attribute is omitted. For example, card 7 consists of VISA, 10%, $5,000.

To take the example you provided in your last post, the row you should have inputted in your data file should look like this instead (I'm using "|" to indicate that we're switching from one attribute to the next):

5 | 0 0 0 | 1 0 0 | 0 1 0 | 0 0 1 | 1 | 0 | 0

Instead of your columns being IV1, IV2, IV3, IV4, IV5, IV6, they should be IV1_2, IV1_3, IV1_4, IV2_2, IV2_3, IV2_4, IV3_2, IV3_3, IV3_4, IV4_2, IV4_3, IV4_4, IV5, IV6, IV7. (the last 3 are binomial already, so you don't need to manipulate them).

Hope this helps.