How could I collect and analyze data for this question/experiment?

#1
Hello everyone,

I want to run an experiment with 4 groups (4 manipulated independent variables). To be precise, I give four different descriptions of an investment portfolio (managed by company X, managed by advisor Y etc.).

Subjects see the development of the portfolio over time and can decide to sell or keep the portfolio at 5 points in time. The portfolio is continuously going down. The dependent variable I measure is "how long do they keep the portfolio?". In the end I want to say something about trust, so how much they need to loose until they loose trust in the portfolio(manager).

I am wondering how I can analyze the data to find out differences between the four groups.

My wild (and probably wrong idea):
Treat the measured DV like a scale. So it's either sold at t1, t2, t3, t4 or t5. I probably cannot assume that it is a Likert scale since Likert scaling assumes distances between each item are equal. This is not the case since the portfolio is not declining linearly.
Can I nonetheless transform it into a scale from 1-5 and just calculate a multivariate ANOVA?

Any other ideas? I did not deploy the survey yet so changes in data collection are possible.

Thanks in advance!
 

Karabiner

TS Contributor
#2
I probably cannot assume that it is a Likert scale
Correct. A Likert scale is an instrument for the measurement of attitutes or opinions, and it is the sum of responses on several Likert-items. None of these characteristics is apparent here.
Can I nonetheless transform it into a scale from 1-5 and just calculate a multivariate ANOVA?
What do you mean by multivariate ANOVA? This would mean, you want to analyse several dependent variables. But seemingly you have just 1 dependent variable (point in time).

Or, do you mean a multi-factorial ANOVA, i.e. ANOVA with several (more than one) factors? If yes, what are these factors, in addition to group membership?

If you just have 1 factor (group membership) you could simply use Kruskal-Wallis H-test, which is for ordinal scaled dependent variables.

With kind regards

Karabiner
 
#3
Thanks Karabiner!

You are right, neither mutlivariate nor multi-factorial ANOVA seems to be appropriate.

Kruskal-Wallis H-test sounds appropriate. I have ordinal data as long as the worth of the portfolio keeps declining with every point in time, right? Which test would I have to use if that is not the case? I mean if the portfolio is worth 1000€ at day one, 1100€ at day 2 and 900€ at day 3 for example.

Can you say anything about the robustness and is it common practice to use this test in experiments?

I really appreciate your advice!
 

Karabiner

TS Contributor
#4
I have ordinal data as long as the worth of the portfolio keeps declining with every point in time, right?
You have ordinal data, since t5 > t4 > t3 > t2 > t1. You wrote "The dependent variable I measure is 'how long do they keep the portfolio?' "
Which test would I have to use if that is not the case? I mean if the portfolio is worth 1000€ at day one, 1100€ at day 2 and 900€ at day 3 for example.
I am afraid I do not understand the question. Is the experimental design dfferent from what you initially described? You wrote "The portfolio is continuously going down."

With kind regards

Karabiner
 
#5
Sorry if I am confusing you.

So the respondent sees the following:

January 1 (t1)
Your portfolio's worth: 1000€
Do you want to sell? yes/no

February 2 (t2)
Your portfolio's worth: 900€
Do you want to sell? yes/no

March 1 (t3)
Your portfolio's worth: 850€
Do you want to sell? yes/no

April 1 (t4)
Your portfolio's worth: 780€
Do you want to sell? yes/no

May 1 (t5)
Your portfolio's worth: 780€
Do you want to sell? yes/no

As soon as they answer yes once, they won't see the subsequent questions/months.

It is always declining so t5 > t4 > t3 > t2 > t1 is true. This means I can use Kruskal-Wallis H-test, right?

Just as a thought experiment, I was wondering what test I would use if the portfolio is going up and down. I am sure there are many comparable experiments in Economics.