Effects coding

[prefer]

Dear all,

I have produced a design for a pilot survey and and data is already collected. Now I want to enter the data in excel.

For example we have nut.effects[0.734|0.588]*nutrition[1,2,3]. Now when the level 1 is applicable in an alternative, it takes the value of 1 otherwise 0, when the level 2 is applicable in an alternative, it takes the value 1 otherwise 0, etc. As usual, I don't want to estimate the parameter for the base level as it is a reference point, so it is left out in the data but it takes the value of -1 anyway. Am I correct?

And even if we use effects coding in Ngene, the real data appears to be dummy, i.e. 0 and 1. Am I correct? I thank you very much for your time.

Mohammed
PhD candidate
University of Copenhagen

[Michiel Bliemer]

No that is not correct. You include all levels in the data, you do not leave out the reference level.

Suppose that your utility function looks like U = ... nut1 * nutrition1 + nut2 * nutrition2, where nutrition1 is the attribute for the first level, and nutrition2 is the attribute for the second level, and nut1 and nut2 are parameters (with priors 0.734 and 0.588). The third level is the base level.

So suppose that in your data you have levels 1, 2, and 3 appearing, then in the data you will find the following:
nutrition1, nutrition2
1,0
0,1
-1,-1

So you explicitly put the third level in the data (using -1,-1 in case of effects coding), you cannot simply remove this level from the data.

[prefer]

Hi Michiel,

Thank you very much for your response. I get your message but even if they are in the data we don't estimate a parameter for the base level which is denoted as -1,-1, right?

Kind regards
Mohammed
PhD candidate
University of Copenhagen

[Michiel Bliemer]

That is correct, you only estimate 2 parameters if you have 3 levels. All three levels are representing in the coding of the two attribute levels.

[prefer]

Ok Michiel but I am a little bit confused. Why do we include the - 1 in the data if we don't estimate a parameter for them. For instance, this is what I did when I enter the data to excel: Imagine the following choice set

Option 1 Option 2
Nutritional value Average Low
Food safety control Very high Standard
Recommendation None Friends and relatives
Product location Big supermarkets Local or street markets
Price 50 120
Which option do you choose?

Nutritional value has three levels as high, average and low, so when I enter the data for the above choice set, I did this:

Nut1_high nut1_average Nut2_high Nut2_average
0 1 0 0

Nut1 is in alternative 1 and Nut2 is in alternative 2. As you can see, no -1 is included- Am I not correct? I thank you for your help!

[Michiel Bliemer]

See my example I provided above, the base level does influence the parameter estimates of the first and second level. The levels -1, -1 are set for the first two levels, hence impacting the parameter estimates. If you set them to 0, 0 (as you do with dummy coding) you get different parameter estimates.

So if you have the following Nutritional value in the choice task you gave: Average Low

And Nutritional value has three levels: high, average and low, where I assume that low is the base, then you need to put this in the data for Average, Low:

Nut1_high nut1_average Nut2_high Nut2_average
0 1 -1 -1

[prefer]

Dear Michiel,

I really thank you very much - you really help me resolve this problem which otherwise would be catastrophic

Kind regards
Mohammed