Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • xtreg fixed effects

    Hello Everyone!

    I have a very basic question regarding fixed effects which is bothering me.

    I have a dataset that has a tsset on school and year.

    The model that I am running is a simple one:

    Code:
    xtreg total_enr drink_water toilets boundary_wall computer sports library main_gate sewerage electricity play_ground, robust
    Where the dependent variable is total enrollment and the remainder variable are all binary

    Suppose I wanted to add district and time fixed effects. The regression then (I believe) will become:

    Code:
    xtreg total_enr drink_water toilets boundary_wall computer sports library main_gate sewerage electricity play_ground i.dist_id i.year, fe robust
    Q 1. Will this mean that I have 3 fixed effects in this model? One for school, district and year?
    Q 2. What if I did not want school fixed effects but rather just required district and year fixed effects? Can i simply run the same regression above without the fe option? (This doesnt make sense to me)

    Q 3. The other thing I was thinking of doing in order to run a district and year fixed effects only was to tsset by district and add a i.year to the model with a ,fe option? Does this make more sense or the second option is more practical?

    Code:
    * Example generated by -dataex-. To install: ssc install dataex
    clear
    input int dist_id long emiscode int year float total_enr byte(drink_water toilets boundary_wall computer sports library main_gate sewerage electricity play_ground)
    311 31110001 2004  935 1 1 1 1 1 1 1 0 1 1
    311 31110001 2005 1032 1 1 1 1 1 1 1 0 1 1
    311 31110001 2006 1316 1 1 1 1 1 1 1 0 1 1
    311 31110001 2007 1450 1 1 1 1 1 1 0 0 1 1
    311 31110002 2004 1009 1 1 1 1 1 1 1 1 1 0
    311 31110002 2005 1145 1 1 1 1 1 1 1 1 1 0
    311 31110002 2006 1050 1 1 1 1 1 1 1 1 1 0
    311 31110002 2007 1068 1 1 1 1 1 1 1 1 1 0
    311 31110003 2004  571 1 1 1 0 0 1 1 1 1 0
    311 31110003 2005  620 1 1 1 0 0 1 1 1 1 0
    311 31110003 2006  739 1 1 1 0 0 1 1 1 1 0
    311 31110003 2007  805 1 1 1 0 1 1 1 1 1 0
    311 31110004 2004  467 1 1 1 0 1 1 1 0 1 1
    311 31110004 2005  579 1 1 1 0 1 1 1 0 1 1
    311 31110004 2006  669 1 1 1 0 1 1 1 0 1 1
    311 31110004 2007  624 1 1 1 0 0 1 1 0 1 0
    311 31110005 2004  662 1 1 1 0 1 1 1 0 1 1
    311 31110005 2005  788 1 1 1 0 1 1 1 0 1 1
    311 31110005 2006  943 1 1 1 0 1 1 1 0 1 1
    311 31110005 2007  938 1 1 1 0 1 1 1 0 1 1
    311 31110006 2004  388 1 0 1 0 0 1 1 0 1 0
    311 31110006 2005  435 1 0 1 0 0 1 1 0 1 0
    311 31110006 2006  425 1 0 1 0 0 1 1 0 1 0
    311 31110006 2007  409 1 0 1 0 0 1 1 0 1 0
    311 31110007 2004  306 1 0 0 0 1 1 1 0 1 1
    311 31110007 2005  359 1 0 0 0 1 1 1 0 1 1
    311 31110007 2006  411 1 0 0 0 1 1 1 0 1 1
    311 31110007 2007  437 1 0 0 0 1 1 1 0 1 1
    311 31110008 2004  410 1 1 1 0 0 1 1 0 1 0
    311 31110008 2005  507 1 1 1 0 0 1 1 0 1 0
    311 31110008 2006  568 1 1 1 0 0 1 1 0 1 0
    311 31110008 2007  593 1 1 1 0 1 1 1 0 1 0
    311 31110009 2004  279 1 1 1 0 0 1 1 0 1 0
    311 31110009 2005  299 1 1 1 0 0 1 1 0 1 0
    311 31110009 2006  358 1 1 1 0 0 1 1 0 1 0
    311 31110009 2007  499 1 1 1 0 0 1 1 0 1 0
    311 31110010 2004  777 1 1 1 0 1 1 1 1 1 1
    311 31110010 2005  824 1 1 1 0 1 1 1 1 1 1
    311 31110010 2006  861 1 1 1 0 1 1 1 1 1 1
    311 31110010 2007  858 1 1 1 0 0 1 1 1 1 1
    311 31110011 2004  377 1 0 1 0 1 1 0 0 1 1
    311 31110011 2005  436 1 0 1 0 1 1 0 0 1 1
    311 31110011 2006  564 1 0 1 0 1 1 0 0 1 1
    311 31110011 2007  453 1 0 0 0 1 1 0 0 1 1
    311 31110012 2004  314 1 1 1 0 1 1 1 1 1 1
    311 31110012 2005  407 1 1 1 0 1 1 1 1 1 1
    311 31110012 2006  531 1 1 1 0 1 1 1 1 1 1
    311 31110012 2007  618 1 1 1 0 1 1 1 1 1 1
    311 31110013 2004  439 1 1 0 0 1 1 1 1 1 1
    311 31110013 2005  574 1 1 0 0 1 1 1 1 1 1
    311 31110013 2006  611 1 1 0 0 1 1 1 1 1 1
    311 31110013 2007  577 1 1 1 0 1 1 1 1 1 1
    311 31110014 2004  479 1 1 0 0 1 1 1 0 1 1
    311 31110014 2005  588 1 1 0 0 1 1 1 0 1 1
    311 31110014 2006  663 1 1 0 0 1 1 1 0 1 1
    311 31110014 2007  698 1 1 1 0 0 1 1 0 1 0
    311 31110015 2004  274 1 0 0 0 1 1 1 0 1 1
    311 31110015 2005  309 1 0 0 0 1 1 1 0 1 1
    311 31110015 2006  382 1 0 0 0 1 1 1 0 1 1
    311 31110015 2007  384 1 1 0 0 0 1 0 0 1 0
    311 31110017 2004  358 1 1 0 0 1 1 1 0 1 1
    311 31110017 2005  377 1 1 0 0 1 1 1 0 1 1
    311 31110017 2006  457 1 1 0 0 1 1 1 0 1 1
    311 31110017 2007  436 1 1 0 0 1 1 1 0 1 1
    311 31110018 2004  900 1 1 1 0 1 1 1 0 1 0
    311 31110018 2005  914 1 1 1 0 1 1 1 0 1 0
    311 31110018 2006  927 1 1 1 0 1 1 1 0 1 0
    311 31110018 2007  786 1 1 1 0 1 1 1 0 1 0
    311 31110019 2004  493 1 1 1 0 0 1 1 1 1 0
    311 31110019 2005  559 1 1 1 0 0 1 1 1 1 0
    311 31110019 2006  543 1 1 1 0 0 1 1 1 1 0
    311 31110019 2007  451 1 1 1 0 0 1 1 1 1 0
    311 31110020 2004 1659 1 1 1 0 1 1 1 1 1 1
    311 31110020 2005 1766 1 1 1 0 1 1 1 1 1 1
    311 31110020 2006 1827 1 1 1 0 1 1 1 1 1 1
    311 31110020 2007 1619 1 1 1 0 1 1 1 1 1 1
    311 31110021 2004 1139 1 1 1 1 1 1 1 0 1 1
    311 31110021 2005 1413 1 1 1 1 1 1 1 0 1 1
    311 31110021 2006 1282 1 1 1 1 1 1 1 0 1 1
    311 31110021 2007 1344 1 1 1 1 1 1 1 0 1 1
    311 31110022 2004  377 1 1 1 0 0 0 1 0 1 0
    311 31110022 2005  396 1 1 1 0 0 0 1 0 1 0
    311 31110022 2006  390 1 1 1 0 0 0 1 0 1 0
    311 31110022 2007  377 1 1 1 0 0 0 1 0 1 0
    311 31110023 2004  362 1 1 0 0 1 1 0 0 1 1
    311 31110023 2005  441 1 1 0 0 1 1 0 0 1 1
    311 31110023 2006  533 1 1 0 0 1 1 0 0 1 1
    311 31110023 2007  594 1 1 0 0 1 1 0 0 1 1
    311 31110024 2004  448 1 1 0 0 0 1 0 1 1 0
    311 31110024 2005  435 1 1 0 0 0 1 0 1 1 0
    311 31110024 2006  438 1 1 0 0 0 1 0 1 1 0
    311 31110024 2007  462 1 1 0 0 1 1 1 1 1 0
    311 31110025 2004  382 1 1 1 0 1 1 1 0 1 0
    311 31110025 2005  371 1 1 1 0 1 1 1 0 1 0
    311 31110025 2006  361 1 1 1 0 1 1 1 0 1 0
    311 31110025 2007  337 1 1 1 0 1 1 1 0 1 0
    311 31110026 2004  381 1 1 1 0 1 1 1 1 1 1
    311 31110026 2005  402 1 1 1 0 1 1 1 1 1 1
    311 31110026 2006  458 1 1 1 0 1 1 1 1 1 1
    311 31110026 2007  487 1 1 1 0 1 1 1 1 1 1
    end
    label values drink_water yesno0
    label values sports yesno0
    label def yesno0 1 "Yes", modify
    label def yesno0 0 "No", modify
    label values toilets yesno
    label values boundary_wall yesno
    label values computer yesno
    label values library yesno
    label values main_gate yesno
    label values sewerage yesno
    label values electricity yesno
    label values play_ground yesno
    label def yesno 0 "No", modify
    label def yesno 1 "Yes", modify

  • #2
    Q 1. Will this mean that I have 3 fixed effects in this model? One for school, district and year?
    If schools don't change districts over the sample period, then the district dummies will be collinear with the school fixed effects and these will drop out.

    Q 2. What if I did not want school fixed effects but rather just required district and year fixed effects? Can i simply run the same regression above without the fe option? (This doesnt make sense to me)
    With -xtreg, fe-, once you specify a panel identifier using xtset, you cannot subsequently exclude this variable. On the other hand, you will have an error of multiple observations within years if you try to xtset using district and time. Therefore, you can switch to different estimators or not include a time variable in the xtset command. If you run xtreg without the -fe- option, you will be estimating a random effects model which is the default, so you don't want to do this.

    Code:
    regress y x... i.year, a(district)
    areg y x... i.year, a(district)
    xtset district
    xtreg y x... i.year, fe
    *ssc install reghdfe
    reghdfe y x..., a(district year)
    Q 3. The other thing I was thinking of doing in order to run a district and year fixed effects only was to tsset by district and add a i.year to the model with a ,fe option? Does this make more sense or the second option is more practical?
    Yes, exactly. That is one of my suggestions above.

    Comment


    • #3
      Thank you so much for clearing things out, will be great if more light can be shed on the topic as well or possibly more resources will be helpful!

      Comment


      • #4
        Note, that with fixed effects, you are estimating the parameters based on variation within panels. So, your effect of sewers will only be the effect if sewers switches or changes within a panel. While it is reasonably common to assume that the effect of such switches is the same as the panel effect (the Hausman test makes this assumption), it is not necessarily correct. That is, the effect of going from no sewer to sewer on something may be different than the effect of having sewers throughout or not having sewers throughout. Note also if you want to generalize to the effect of having sewers throughout versus not than the within effect from fixed effects estimates may not be the right parameter.

        Comment


        • #5
          Originally posted by Phil Bromiley View Post
          Note, that with fixed effects, you are estimating the parameters based on variation within panels. So, your effect of sewers will only be the effect if sewers switches or changes within a panel. While it is reasonably common to assume that the effect of such switches is the same as the panel effect (the Hausman test makes this assumption), it is not necessarily correct. That is, the effect of going from no sewer to sewer on something may be different than the effect of having sewers throughout or not having sewers throughout. Note also if you want to generalize to the effect of having sewers throughout versus not than the within effect from fixed effects estimates may not be the right parameter.
          thank you for the reply, makes sense!

          Comment

          Working...
          X