Crossposted with Stack Exchange Economics: https://economics.stackexchange.com/...ut-independent
Dear Statalist,
I have a conceptual question regarding using the t-test via ttest, by(group).
I have experimental data where participants (two different types: students and professionals (mutually exclusive - there is no overlap between them) participate in an investment experiment. Their first task is to create a hypothetical portfolio for a hypothetical investor, and the second task is to manage the portfolio using one of three options. The treatment condition is the provision of information - in the treatment condition, I provide additional information regarding the consequences of the selected action (one of the 3), and in the control group, this information is absent.
Now, my supervisor asked that I conduct a t-test on the equality of the means regarding the experimental portfolio actions by participant type. The goal is to find out whether there is a difference between the participant types (student vs. professional) who selected the first portfolio management option (and the second and third). First, I checked that the variances between the groups are equal. Second, I did ttest experimental action 1, by(group). I find statistically significant and interesting results. I repeated the t-tests for the second and third actions, respectively.
However, the internet has now cautioned me against the use of the t-test because I (might) have repeated observations. To that: Yes, every participant engages in four rounds: so four experimental actions (choice of three) per participant. However, since I analyze the three actions separately - I am not sure if these are really repeated actions because a repeated action would mean that the same participant chooses the first action more than once - but I also could not think of a way to count(?) how often the same participant ID chooses a particular action?
On the internet, I found information regarding a paired t-test, but I don't think this is appropriate here because there is no overlap between the participant groups - it is not an initial measurement - intervention (drug provision, etc.) - the last measurement of the same individual.
I am also not sure regarding the independence (or absence thereof) because each round in the experiment provides completely different information regarding the hypothetical investor for whom the portfolio is constructed and managed - so (I argue) that the rounds are not linked to each other - and might be seen as independent (?) Yes, the task is the same: construct and manage a portfolio, but the information provided on the basis of which to construct and manage the portfolio is entirely different.
Perhaps, as an addition, in this t-test, I am not interested in the effect of the treatment condition on the experimental actions - here, I am only interested in whether there is a difference between the groups with respect to each of the three possible experimental actions.
I would be most grateful for enlightenment as to whether I can use the t-test for my analysis.
P.S. I also do OLS (LPM) (and logit) regressions, where I find no significance between the participant type with respect to each of the 3 portfolio management options, but I really want to gain more insights into the means of the respective groups in each of the 3 actions.
Dear Statalist,
I have a conceptual question regarding using the t-test via ttest, by(group).
I have experimental data where participants (two different types: students and professionals (mutually exclusive - there is no overlap between them) participate in an investment experiment. Their first task is to create a hypothetical portfolio for a hypothetical investor, and the second task is to manage the portfolio using one of three options. The treatment condition is the provision of information - in the treatment condition, I provide additional information regarding the consequences of the selected action (one of the 3), and in the control group, this information is absent.
Now, my supervisor asked that I conduct a t-test on the equality of the means regarding the experimental portfolio actions by participant type. The goal is to find out whether there is a difference between the participant types (student vs. professional) who selected the first portfolio management option (and the second and third). First, I checked that the variances between the groups are equal. Second, I did ttest experimental action 1, by(group). I find statistically significant and interesting results. I repeated the t-tests for the second and third actions, respectively.
However, the internet has now cautioned me against the use of the t-test because I (might) have repeated observations. To that: Yes, every participant engages in four rounds: so four experimental actions (choice of three) per participant. However, since I analyze the three actions separately - I am not sure if these are really repeated actions because a repeated action would mean that the same participant chooses the first action more than once - but I also could not think of a way to count(?) how often the same participant ID chooses a particular action?
On the internet, I found information regarding a paired t-test, but I don't think this is appropriate here because there is no overlap between the participant groups - it is not an initial measurement - intervention (drug provision, etc.) - the last measurement of the same individual.
I am also not sure regarding the independence (or absence thereof) because each round in the experiment provides completely different information regarding the hypothetical investor for whom the portfolio is constructed and managed - so (I argue) that the rounds are not linked to each other - and might be seen as independent (?) Yes, the task is the same: construct and manage a portfolio, but the information provided on the basis of which to construct and manage the portfolio is entirely different.
Perhaps, as an addition, in this t-test, I am not interested in the effect of the treatment condition on the experimental actions - here, I am only interested in whether there is a difference between the groups with respect to each of the three possible experimental actions.
I would be most grateful for enlightenment as to whether I can use the t-test for my analysis.
P.S. I also do OLS (LPM) (and logit) regressions, where I find no significance between the participant type with respect to each of the 3 portfolio management options, but I really want to gain more insights into the means of the respective groups in each of the 3 actions.
