[Carlo Lazzaro]

Guest:
thanks for providing further details (that seems to change your research goal the way I got it from your original post).
Hence, if you do not have panel anymore, OLS is the way to go.
If that were the case, Alfonso is right in highlighting that, being no repeated measures on the same unit, -cluster- is off target there.

[Guest]

What is a wave? I see from the code that basically the wave is the time component, right?

yes in total there are 8 waves= 8 years and im using panel data

The question you're asking is about parents who change their relationship, right? However, if you just take one wave (the 8th one) that is comparing parents that are in a relationship to parents that are out of a relationship. You don't observe the same couple going into the relationship and out of it, because you don't allow time to change. I'm not sure this is measuring the same thing.

in the fe models i can't compare people who changed their relationship status cause there are only people in the model who changed (within). but to test the hypothese H1: Parents who change their relationship status have a lower Parent-Child-Relationshipquality than parents who are in a stable relationship I need a between model. So I use the value of the parent-child-relationship-quality as my dependent variable of the last year (wave). The independent variable is 0 no change and 1 change. I had to recode the variable because otherwise it would only say if there was a change from wave 7 to 8. But I need to know if there was a change in general. For this I counted all changes of the IDs. E.g. for the ID 003 there was a change in wave 4 and 6 but not in wave 8 . I don't know if this is the right way to do and furthermore if I gen the variable correctly.

I am very confused about everything in this analysis.

Me too, and I have to hand in me thesis in 3 weeks

[Guest]

Guest:
thanks for providing further details (that seems to change your research goal the way I got it from your original post).
Hence, if you do not have panel anymore, OLS is the way to go.
If that were the case, Alfonso is right in highlighting that, being no repeated measures on the same unit, -cluster- is off target there.

Yes I tried to explain it a bit better above. I'm really really confused right now and have no idea how to handle this.

[Alfonso Sánchez-Peñalver]

Hi Guest,

let's concentrate on the variables a little more. I'm still a bit confused after your explanation.

Ok so you have different households for 6 years (waves 3 to 8). So far so good, you have a panel. I have looked at the data you originally posted and I get the following

. xtsum transition

Variable         |      Mean   Std. Dev.       Min        Max |    Observations
-----------------+--------------------------------------------+----------------
transi~n overall |       .05   .2190429          0          1 |     N =     100
         between |             .1323341          0         .5 |     n =      29
         within  |              .168175       -.45        .85 | T-bar = 3.44828

This shows me that you have both within and between variation in transition. Now, is transition an indicator of whether a couple is in a relationship (1) or they have split up (0), or the other way around? If so you do have the information about whether a couple split up or not.

I also do

. xtsum bezqual_kind

Variable         |      Mean   Std. Dev.       Min        Max |    Observations
-----------------+--------------------------------------------+----------------
bezqua~d overall |  3.721667   .6321249   1.666667   4.833333 |     N =     100
         between |              .577732        2.5   4.555556 |     n =      29
         within  |             .3863493   2.721667   4.888333 | T-bar = 3.44828

which also tells me there is both between and within variation in your dependent variable. Because of this, I agree with Carlo Lazzaro when he says that he doesn't agree with your advisor's advice.

Now, fixed effects captures the within effect: that is the effect of a couple entering or exiting a relationship, if the unit of observation is the couple. It removes all between effect. That actually addresses your H1a: Parents who change their relationship status have a lower Parent-Child-Relationshipquality than parents who are in a stable relationship.

I am still confused about why you would only use the value for the 8th wave in bezqual_kind. I don't see the logic of doing that. Let me know if I have missed something so far.