[Warner de Jong]

Dear Carlo,

Do you mean "Example 4: Growth-curve model with correlated random effects"?

[Carlo Lazzaro]

Warner:
no.
See page 376 of the abovementioned entry. The example #4 starts tecnically with: use http://www.stata-press.com/data/r14/productivity.
By the way; are you using Stata 14?

[Warner de Jong]

Carlo:
The link does not seem to work.
And, yes, version 14.2

EDIT 1: http://www.stata-press.com/data/r14/ does work. I can choose from a variety of manuals, I'm guessing it is this one: Multilevel Mixed-Effects Reference Manual [ME]

EDIT 2: I do see that http://www.stata-press.com/data/r14/productivity.dta exists as a Stata file

EDIT 3: I'm looking for these links using internet explorer, perhaps you are using Stata?

EDIT 4: I found it under http://www.stata-press.com/manuals/m...erence-manual/

[Carlo Lazzaro]

Warner:
my mistake, sorry for this.
I should have reported:

 use http://www.stata-press.com/data/r14/productivity

which works for me (copied and pasted in Stata 14.2):

. use http://www.stata-press.com/data/r14/productivity
(Public Capital Productivity)

The explanation of the worked example in -mixed- entry, Stata 14.2 .pdf manual (page 376) is good (the underlying theory is unavoidably challenging, though).
To spot the right manual for -mixed-, simply type -help mixed- from within Stata, click on the -Also see- option, scroll it down, click again on the -mixed- manual (.pdf icon) and you will directed to the -mixed- entry..

[Warner de Jong]

After checking this data set I see that each state has multiple entries.
In my case, each firm has only one entry. Namely, the year in which the previous CEO was succeeded by the current CEO.

I guess that if I would use a panel data set than I would need to attach an ID for every CEO, otherwise I could not tell when a succession would take place.

Or could I possibly keep the dataset that I currently have?

If need be, I could share the dataset without the firm names if that would make it easier.

[Carlo Lazzaro]

Warner:
multiple observations for the same -id- are not a prerequisite for mixed model.
You can go -mixed- even with one wave of data only (as in your case), but you have to check for the overarching/nesting structure of your data (CEO nested within firms nested within industries).

[Warner de Jong]

Carlo:
I was wondering on how I should put the variables into the equation, could you help me with that?

The variables that I use are as follows

describe
obs: 120
vars: 13 30 May 2017 19:25
size: 4,920
--------------------------------------------------------------------------------------------------------------------------------------------------------
storage display value
variable name type format label variable label
--------------------------------------------------------------------------------------------------------------------------------------------------------
Company_ID _float _%9.0g
Year _______int ___%10.0g Year of CEO succession
Boardsize __byte __%10.0g Board size
Boardtype __byte __%10.0g 1 = inside, 0 = outside
CEOtype ____byte __%10.0g 1 = follower, 2 = contender, 3 = outsider
Duality ____byte __%10.0g 1 = Duality, 0 = not
PC_chairman_byte __%10.0g 1 = Prev. CEO is chair, 0 = not
PC_duality _byte __%10.0g 1 = Prev. CEO had duality, 0 = not
Pre_ROA ____double %10.0g Pre-succession performance (3-year ROA)
Post_ROA ___double %10.0g Post-succession performance (3-year ROA)
SIC2 _______byte __%10.0g SIC 2-digit industry code of 37 different industries, some industries have 1 firm
Ind_ROA ____double %10.0g Industry performance (3-year ROA) from averages
LogSales ___float _%9.0g Log sales for firm size
-------------------------------------------------------------------------------------------------------------------------------------------------------

I thought of writing it down as this, the firm would then be nested in the industry. I did not nest the CEO in the firm as I wish to investigate the effect of each individual CEO type.

mixed Post_ROA Pre_ROA Boardsize LogSales Ind_ROA i.PC_chairman i.PC_duality i.Duality i.Year i.CEOtype i.Boardtype || SIC2: || ID:

Or should it be that I immediately look for the hypothesized effects. Namely that CEO type interacts with Board type?
mixed Post_ROA Pre_ROA Boardsize LogSales Ind_ROA i.PC_chairman i.PC_duality i.Duality i.Year i.SIC2 || Boardtype: || CEOtype:

[Carlo Lazzaro]

Warner:
I would try your first -mixed- code.
However, -mixed- does not solve the risk of heterogeneity in your OLS (ie, the fixed part of your -mixed- model): I would search for your supervisor's opinion on that.

[Warner de Jong]

Carlo:
Thank you very very very much for all of your help.

I will address the issue with my supervisor.

Warner