- Standard question types
- Advanced question types
- Multiple choice question type
- Text question- comment box
- Matrix multi-point scales question type
- Rank order question
- Smiley-rating question
- Image question type
- Date and time question type
- CAPTCHA question type
- Net Promoter Score question type
- Van Westendorp's price sensitivity question
- Choice modelling questions
- Side-By-Side matrix question
- Homunculus question type
- Randomizer
- Block rotation using randomization
- Predictive answer options
- Presentation text questions
- Multiple choice: select one
- Multiple choice: select many
- Page timer
- Contact information question
- Matrix multi-select question
- Matrix spreadsheet question
- Closed card sorting question
- Flex Matrix
- Text Slider Question Type
- Graphical Rating Scales
- Rank Order - Drag and Drop
- Bipolar Matrix - Slider
- Bipolar Matrix Likert Scale

- Answer type
- Reorder questions
- Question tips
- Text box next to question
- Text question settings
- Adding other option
- Matrix question settings
- Image rating question settings
- Scale options for numeric slider question
- Constant sum question settings
- Budget scale question settings
- Setting default answer option
- Exclusive option for multiple choice questions
- Validate question
- Bulk validation
- Remove validation message
- Question separators
- Question Code
- Page breaks in survey
- Survey introduction with acceptance checkbox
- RegEx Validation
- Question Library
- Embed Media
- Slider Start Position

- Branching - Skip Logic
- Compound Branching
- Compound or delayed branching
- Dynamic quota control
- Dynamic text or comment boxes
- Extraction logic
- Show or hide question logic
- Dynamic show or hide
- Scoring logic
- Net promoter scoring model
- Piping text
- Survey chaining
- Looping logic
- Branching to terminate survey
- Logic operators
- Selected N of M logic
- JavaScript Logic Syntax Reference
- Block Flow
- Block Looping

- Survey settings
- Save & continue
- Anti Ballot Box Stuffing (ABBS) - disable multiple responses
- Deactivate survey
- Admin confirmation email
- Action alerts
- Survey timeout
- Finish options
- Spotlight report
- Print survey response
- Search and replace
- Compact view
- Survey Timer
- Allowing multiple respondents from the same device
- Text Input Size Settings

- Live survey URL
- Customize survey URL
- Create email invitation
- Personalizing emails
- Email invitation settings
- Email list filter
- Survey reminders
- Export batch
- Email status
- Spam index
- Send surveys via SMS
- Phone & paper
- Adding responses manually
- SMS Pricing
- Embedding Question In Email
- Deleting Email Lists
- Multilingual Survey Distribution

- Conjoint analysis designs
- Conjoint part worths calculation
- Conjoint calculations and methodology
- Conjoint attribute importance
- Conjoint profiles
- Market segmentation simulator
- Conjoint brand premium and price elasticity
- What is MaxDiff scaling
- MaxDiff settings
- Anchored MaxDiff Analysis
- MaxDiff FAQ
- MaxDiff- Interpreting Results

Calculating Part-Worths Values:

We use the following algorithm to calculate CBC Conjoint Part-Worths:

**NOTATION**Let there be R respondents, with individuals r = 1 ... R

Let each respondent see T tasks, with t = 1 ... T

Let each task t have C configurations (or concepts), with c = 1 ... C (C in our case is usually 3 or 4)

If we have A attributes, a = 1 to A, with each attribute having La levels, l = 1 to La, then the part-worth for a

particular attribute/level is w’(a,l). It is this (jagged array) of part worths we are solving for in this exercise. We can

simplify this to a one-dimensional array w(s), where the elements are:

{w’(1,1), w’(1,2) ... w’(1,L1), w’(2,1) ... w’(A,LA)} with w having S elements.

A specific configuration x can be represented as a one-dimensional array x(s), where x(s)=1 if the specific

level/attribute is present, and 0 otherwise.

Let Xrtc represent the specific configuration of the cth configuration in the tth task for the rth respondent. Thus the

experiment design is represented by the four dimensional matrix X with size RxTxCxS

If respondent r chooses configuration c in task t then let Yrtc=1; otherwise 0.

**UTILITY OF A SPECIFIC CONFIGURATION**The Utility Ux of a specific configuration is the sum of the part-worths for those attribute/levels present in the configuration, i.e. it is the scalar product x.w

**THE MULTI-NOMIAL LOGIT MODEL**For a simple choice between two configurations, with utilities U1 and U2, the MNL model predicts that configuration 1 will be chosen

EXP(U1)/(EXP(U1) + EXP(U2)) of the time (a number between 0 and 1).

For a choice between N configurations, configuration 1 will be chosen

EXP(U1)/(EXP(U1) + EXP(U2) + ... + EXP(UN)) of the time.

**MODELED CHOICE PROBABILITY**Let the choice probability (using MNL model) of choosing the cth configuration in the tth task for the rth respondent be:

Prtc=EXP(xrtc.w)/SUM(EXP(xrt1.w), EXP(xrt2.w), ... , EXP(xrtC.w))

**LOG-LIKELIHOOD MEASURE**The Log-Likelihood measure LL is calculated as:

Prtc is a function of the part-worth vector w, which is the set of part-worths we are solving for.

**SOLVING FOR PART-WORTHS USING MAXIMUM LIKELIHOOD**We solve for the part-worth vector by finding the vector w that gives the maximum value for LL. Note that we are solving for S variables.

This is a multi-dimensional non-linear continuous maximization problem, and requires a standard solver library. We use the Nelder-Mead Simplex Algorithm.

The Log-Likelihood function should be implemented as a function LL(w, Y, X), and then optimized to find the vector w that gives us a maximum. The responses Y, and the design X are given, and constant for a specific optimization. Initial values for w can be set to the origin 0.

The final part-worths w are re-scaled so that the part-worths for any attribute have a mean of zero, simply by subtracting the mean of the part-worths for all levels of each attribute.

All the Conjoint related tools can be accessed by clicking on:

**Login » Surveys » Reports » Choice Modelling » Conjoint Analysis**