STATISTIK-FORUM.de

[Benni]

Hallo liebe Leute,

ich habe mal eine Frage, die ich auch schonmal in anderem Kontext formuliert habe (in einem englischen Forum, wo mir aber keiner geantwortet hat)

Ich stell hier daher die selbe Frage, zunächst ohne zu übersetzen. Falls jemand die Frage lieber auf deutsch lesen möchte, übersetze ich gerne.

An der Theorie hinter der Frage bin ich sehr interessiert, ich weiß nur nicht, wo ich anfangen soll zu lesen, vielleicht kann mir jemand Hinweise geben. Dank dafür!

Hier die Frage:

For my Bachelor-Thesis I have to assess the link between a firm's connectedness and success.
For this purpose I was given data from the video game industry, which contain information on 2700 games (# of observations). I have to find a link between a developing forms connectedness and the critics score of the games it produced.

I computed a variable which measures the amount of publishing firms a developer has worked with for each game, at the time of the games release, weighted with the overall games the developer has worked with.
My Model is: Critics score (DV) = connectedness of developer * B1 + many controls.
My supervisor told me to control for fixed effects by including n-1 dummy variables for each developing firm (1100), publisher (558), year (18), month (12).

1. Does it make sense to include 1700 dummies, when the overall data only conprise of 2700 observations and the majority of developers / publishers dummies only have 1-2 games released? (match only 1-2 observations) ?
2. A model without all the dummies for publishers and developers gives me a p-value of 0.000 for my coefficient on connectedness, which makes me happy. The "messy" model, which is supposed to check for firm fixed effects, gives me 0.279, which would mean my variable is insignificant. How can I tell which model to trust? Which model would you take? Do you need more information? R² on the many dummy model obviously is extremely high, at 70% as opposed to 1% or less of the other model.

3. How can I find information on model quality and fit in interaction with the number of variables included in the model? I dont know the keywords

Schönes sonniges Wochenende noch,
Gruß,
Benjamin

[Benni]

kann mir jemand Hinweise geben?

[DHA3000]

Ist dir klar, dass du mit einem Panel-Datensatz arbeitest?
Hast du dich damit einmal ein wenig auseinander gesetzt bzw. Ahnung von Panels?

Erläuter einmal genauer, wie sich deine Daten zusammensetzen. Insbesondere die Dimension der einzelnen Variablen, genauso wie die Kontrollvariablen.