New on SSC: CONJOINT: a module to analyse and visualise conjoint (factorial) experiments

[Michael Frith]

With thanks as always to Kit Baum, a new package called -conjoint- is now available on the SCC.

conjoint and is a relatively simple wrapper for Stata (and coefplot) functions that creates a dedicated and straightforward command for analysing and visualising conjoint (factorial) experiments as commonly used in fields such as political science.

Specifically, conjoint can estimate average marginal component effects (AMCE) and marginal means (MM) following the methods described in Hainmueller et al., (2014) and Leeper et al., (2020) and implemented in the R packages, cjoint (Barari et al., 2018) and cregg (Leeper and Barnfield, 2020). conjoint can estimate these for fully randomised designs and AMCEs for designs with unlimited and complex profile constraints. conjoint can also calculate estimates across subgroups, with different baselevels (AMCEs only) and null hypothesis values (MMs only). The results can be simply and easily plotted via the graph option using coefplot.

To install, type:

ssc install conjoint

conjoint's help and ancillary files include two examples using immigration (Hainmueller et al., 2013; see also 2014) and refugee return (Ghosn et al., 2021a; see also 2021b) conjoint experiment datasets.




References

Barari, S., Berwick, E., Hainmueller, J., Hopkins, D., Liu, S., Strezhnev, A., & Yamamoto, T. 2018. cjoint: AMCE Estimator for Conjoint Experiments. R package.

Ghosn, F., Chu, T.S., Simon, M., Braithwaite, A., Frith, M.J., & Jandali, J. 2021a. Replication Data for Journey Back Home: Violence, Anchoring, and Refugee Decisions to Return. https://doi.org/10.7910/DVN/UGI0MH

Ghosn, F., Chu, T.S., Simon, M., Braithwaite, A., Frith, M.J., & Jandali, J. 2021b. The Journey Home: Violence, Anchoring, and Refugee Decisions to Return. American Political Science Review, 1–17. https://doi.org/10.1017/S0003055421000344

Hainmueller, J., Hopkins, D.J., & Yamamoto, T. 2013. Replication data for: Causal Inference in Conjoint Analysis: Understanding Multidimensional Choices via Stated Preference Experiments. https://doi.org/10.7910/DVN/THJYQR

Hainmueller, J., Hopkins, D.J., & Yamamoto, T. 2014. Causal Inference in Conjoint Analysis: Understanding Multidimensional Choices via Stated Preference Experiments. Political Analysis, 22(1): 1-30. https://doi.org/10.1093/pan/mpt024

Leeper, T.J., & Barnfield, M. 2020. cregg: Simple Conjoint Tidying, Analysis, and Visualization. R package.

Leeper, T., Hobolt, S., & Tilley, J. 2020. Measuring Subgroup Preferences in Conjoint Experiments. Political Analysis, 28(2): 207-221. https://doi.org/10.1017/pan.2019.30

[Andrea Maier]

Hello

this sounds very interesting. I tried it, and it works fine.

I have a question and a suggestion to make?

Is there an easy way to change the visuals (similar to the R packages) that in the graphs lines with dots for a specific attribute have the same color?


And a suggestion: The package wants you to have exclusively numerical values but treats them as i.variable (as they were nominal) in the regression, am I right? First I would suggest to make the error code when someone uses "conjoint depvar indepvars that are not numerical, estimate(mace) id(ID) to something that says "give me numbers instead of strings" instead of "baselevels() invalid -- invalid numlist".

Also: Is there a way to include variables that were asked in a normal attribute fashion but are actually numeric values (i.e. think about the immigrant example, giving them savings of X dollars). Because we have this dollars variable and like to see how much money respondents are willing to accept instead of having people with e.g. specific language skills.

Kind regards!

[Joe Greenwood-Hau]

Hello all,

Many thanks to Michael (and Kit) for creating the conjoint package, which I am planning to use for a study that is going into the field shortly.

The fieldwork will be across four countries and, in the first instance, I intend to analyse the experimental results separately for each country.

I am also hoping to merge the data from the four countries together and repeat the analysis on the combined data. As such, I am wondering whether there is a way to include country fixed effects in a conjoint analysis model.

Any advice in this regard would be very gratefully received.

All the best,

Joe

[Fahad Mirza]

Thanks Michael Frith for sharing this and we hope to see more updates on the package!