Announcement

Collapse
No announcement yet.
X
Collapse
  • Page of 2
  • Filter
  • Time
  • Show
Clear All
new posts
  • #1

    Multilevel modeling for panel data using mixed command (individual nested within firms)

    Dear Statalist members,

    Long time observer, first time poster here. I learned a great deal from your posts & answers over the years, and I now have a specific question I can’t quite figure out.

    I am trying to do a sort of a multilevel modelling with “mixed” command, using an unbalanced panel dataset. The overall setting is this:
    DV: Firm-level corporate social responsibility performance (rating)
    IV: Individual top management team characteristic, while controlling for CEO’s attribute

    To elaborate, I am trying to see if a certain characteristic of non-CEO top management team members (TMT) will have an impact on firm-level outcome over and beyond that of the CEO. Let’s call the TMT / CEO characteristic “Trait A”.

    The issue is that I need to include the individual TMT member’s Trait A separately into the model, but the outcome is observed at the firm level. So the model needs to include each of the individual TMT member’s Trait A separately into the model, which would be nested within each firm.

    To do that, I have created multiple identifiers within the dataset: ID for individual TMT members, ID for each individual-TMT pair, ID for each firm (and industry and year identifiers as well). The code I used to run the initial model is like the following (variable names are swapped, but the code itself is identical to one I used):

    So, the question is this: Am I running the multilevel models correctly? The model runs and gives out result tables, but I am not sure if I’m specifying the models correctly given what I am trying to test.

    More specifically, some of the confusing I’m having revolves around issues such as: Am I using the correct grouping variables in each of the levels? Should I try to take more levels into account (e.g., industry)? Can I or should I control for the year effect in the way I’m including them in the model now?

    The code I am using currently is this:

    Code:
    mixed firm_outcome i.year i.industry ceo_trait_A ceo_control_1 ceo_control_2 ///
tmt_control_1 tmt_control2 tmt_trait_A || firm_id: || tmt_firm_pair_id:
    firm_outcome: DV
    i.year: year indicator
    i.industry: industry indicator
    ceo_trait_A: controlling for CEO’s Trait A
    ceo_control_1: additional control for relevant CEO attribute
    ceo_control_2: additional control for relevant CEO attribute
    tmt_control_1: controlling for same attribute in individual TMT members with the CEO
    tmt_control_2: controlling for same attribute in individual TMT members with the CEO
    tmt_trait_A: Variable of interest. Individual TMT member’s Trait A
    firm_id: ID for each firm
    tmt_firm_pair_id: ID for each individual TMT member and firm pair (e.g., TMT A-Firm A = 0001, TMT A-Firm B=0002, …)

    The results look like the following:

    Code:
    Performing EM optimization ...

Performing gradient-based optimization: 
Iteration 0:   log likelihood = -49047.968  
Iteration 1:   log likelihood = -49047.832  
Iteration 2:   log likelihood = -49047.832  

Computing standard errors ...

Mixed-effects ML regression                     Number of obs     =    118,607

        Grouping information
        -------------------------------------------------------------
                        |     No. of       Observations per group
         Group variable |     groups    Minimum    Average    Maximum
        ----------------+--------------------------------------------
                   firm_id |      2,648          1       44.8        238
             tmt_firm_pair_id |     29,321          1        4.0         22
        -------------------------------------------------------------

                                                Wald chi2(36)     =    4458.11
Log likelihood = -49047.832                     Prob > chi2       =     0.0000

------------------------------------------------------------------------------------------
         firm_outcome | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------------------+----------------------------------------------------------------
                    year |
                   1993  |     .00556   .0095535     0.58   0.561    -.0131645    .0242845
                   1994  |   .0049174   .0096436     0.51   0.610    -.0139837    .0238184
                   1995  |   .0118385   .0097604     1.21   0.225    -.0072916    .0309686
                   1996  |   .0102722   .0099117     1.04   0.300    -.0091543    .0296987
                   1997  |   .0033965   .0100088     0.34   0.734    -.0162203    .0230133
                   1998  |  -.0056598   .0100506    -0.56   0.573    -.0253586    .0140391
                   1999  |  -.0076267   .0101267    -0.75   0.451    -.0274746    .0122213
                   2000  |  -.0067193    .010276    -0.65   0.513    -.0268599    .0134212
                   2001  |   .0684061   .0098819     6.92   0.000     .0490379    .0877744
                   2002  |   .0720901   .0098876     7.29   0.000     .0527108    .0914695
                   2003  |   .1515418   .0095472    15.87   0.000     .1328296     .170254
                   2004  |   .1528703   .0096387    15.86   0.000     .1339787    .1717618
                   2005  |   .1554403   .0098182    15.83   0.000      .136197    .1746837
                   2006  |   .1579207   .0096669    16.34   0.000     .1389739    .1768675
                   2007  |   .1783125   .0096435    18.49   0.000     .1594115    .1972134
                   2008  |   .1764963   .0096883    18.22   0.000     .1575075    .1954851
                   2009  |   .1761126   .0097561    18.05   0.000     .1569909    .1952343
                   2010  |   .2589668   .0098262    26.35   0.000     .2397079    .2782258
                   2011  |   .2891228   .0098784    29.27   0.000     .2697616    .3084841
                   2012  |   .2920214   .0099677    29.30   0.000     .2724851    .3115578
                   2013  |   .2948403    .010078    29.26   0.000     .2750878    .3145928
                         |
               industry |
                      2  |   .1142829   .1502563     0.76   0.447    -.1802139    .4087798
                      3  |   .1267328   .1636033     0.77   0.439    -.1939237    .4473893
                      4  |   .3235708   .1465069     2.21   0.027     .0364226     .610719
                      5  |   .2442008   .1479205     1.65   0.099    -.0457179    .5341196
                      6  |   .2979823    .152505     1.95   0.051     -.000922    .5968866
                      7  |   .3295068   .1486819     2.22   0.027     .0380957     .620918
                      8  |   .3907518   .1470049     2.66   0.008     .1026275    .6788762
                      9  |   .3193841   .1472104     2.17   0.030     .0308571    .6079111
                     10  |    .272765   .1943922     1.40   0.161    -.1082366    .6537667
                         |
      ceo_trait_A |   .0004901   .0000709     6.92   0.000     .0003513     .000629
    ceo_control_1 |  -.0001374   .0002338    -0.59   0.557    -.0005958    .0003209
    tmt_control_1 |   .0003279   .0001958     1.68   0.094    -.0000557    .0007116
    ceo_control_2 |  -7.14e-06   7.91e-07    -9.03   0.000    -8.69e-06   -5.59e-06
    tmt_control_2 |   5.97e-06   1.78e-06     3.35   0.001     2.48e-06    9.46e-06
      tmt_trait_A |   .0001101    .000061     1.80   0.071    -9.47e-06    .0002296
            _cons |  -.4758617   .1462839    -3.25   0.001     -.762573   -.1891505
------------------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects parameters  |   Estimate   Std. err.     [95% conf. interval]
-----------------------------+------------------------------------------------
firm_id: Identity               |
                  var(_cons) |   .1396965   .0040593      .1319627    .1478836
-----------------------------+------------------------------------------------
tmt_firm_pair_id: Identity         |
                  var(_cons) |   .0377478    .000635      .0365235    .0390132
-----------------------------+------------------------------------------------
               var(Residual) |   .1020949   .0004819      .1011547    .1030438
------------------------------------------------------------------------------
LR test vs. linear model: chi2(2) = 97906.16              Prob > chi2 = 0.0000

Note: LR test is conservative and provided only for reference.
    I also run the model with an interaction term added, as I’m interested in moderating effect of a certain variable that is based on the relationship between the CEO and the non-CEO TMT member. The code looks like the following:

    Code:
    mixed firm_outcome i.year i.industry ceo_trait_A ceo_control_1 ceo_control_2 ///
tmt_control_1 tmt_control2 c.ceo_tmt_relationship##c.tmt_trait_A || firm_id: || tmt_firm_pair_id:
    *ceo_tmt_relationship: Relationship variable specific for each TMT member and the CEO for each firm.

    The result looks like the following:

    Code:
    Performing EM optimization ...

Performing gradient-based optimization: 
Iteration 0:   log likelihood = -49047.392  
Iteration 1:   log likelihood = -49047.255  
Iteration 2:   log likelihood = -49047.255  

Computing standard errors ...

Mixed-effects ML regression                     Number of obs     =    118,607

        Grouping information
        -------------------------------------------------------------
                        |     No. of       Observations per group
         Group variable |     groups    Minimum    Average    Maximum
        ----------------+--------------------------------------------
                   firm_id |      2,648          1       44.8        238
             tmt_firm_pair_id |     29,321          1        4.0         22
        -------------------------------------------------------------

                                                Wald chi2(38)     =    4459.44
Log likelihood = -49047.255                     Prob > chi2       =     0.0000

-------------------------------------------------------------------------------------------------------------
                            firm_outcome | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
--------------------------------------------+----------------------------------------------------------------
                                       year |
                                      1993  |   .0054582   .0095541     0.57   0.568    -.0132675    .0241838
                                      1994  |   .0047473    .009645     0.49   0.623    -.0141566    .0236513
                                      1995  |      .0116    .009763     1.19   0.235    -.0075352    .0307352
                                      1996  |   .0099628   .0099159     1.00   0.315    -.0094721    .0293977
                                      1997  |   .0030361   .0100144     0.30   0.762    -.0165919     .022664
                                      1998  |  -.0059889   .0100554    -0.60   0.551    -.0256971    .0137194
                                      1999  |  -.0079909   .0101324    -0.79   0.430      -.02785    .0118682
                                      2000  |  -.0070383   .0102804    -0.68   0.494    -.0271876     .013111
                                      2001  |   .0681851   .0098844     6.90   0.000     .0488121    .0875581
                                      2002  |   .0718161   .0098913     7.26   0.000     .0524294    .0912028
                                      2003  |   .1512256   .0095522    15.83   0.000     .1325036    .1699475
                                      2004  |   .1525406   .0096443    15.82   0.000      .133638    .1714431
                                      2005  |   .1550779   .0098248    15.78   0.000     .1358218    .1743341
                                      2006  |   .1575569   .0096735    16.29   0.000     .1385971    .1765166
                                      2007  |   .1779621   .0096495    18.44   0.000     .1590493    .1968748
                                      2008  |   .1761081   .0096957    18.16   0.000     .1571049    .1951114
                                      2009  |   .1756895   .0097649    17.99   0.000     .1565507    .1948283
                                      2010  |   .2585263   .0098356    26.28   0.000     .2392489    .2778037
                                      2011  |   .2886516   .0098891    29.19   0.000     .2692694    .3080338
                                      2012  |   .2915076   .0099807    29.21   0.000     .2719459    .3110694
                                      2013  |   .2943222    .010091    29.17   0.000     .2745442    .3141002
                                            |
                                  industry |
                                         2  |    .114184   .1502386     0.76   0.447    -.1802783    .4086462
                                         3  |   .1263998   .1635835     0.77   0.440    -.1942179    .4470175
                                         4  |   .3232903     .14649     2.21   0.027     .0361752    .6104054
                                         5  |    .243951   .1479034     1.65   0.099    -.0459343    .5338364
                                         6  |   .2977231   .1524874     1.95   0.051    -.0011467    .5965929
                                         7  |   .3292493   .1486642     2.21   0.027     .0378728    .6206257
                                         8  |   .3905855   .1469876     2.66   0.008     .1024951     .678676
                                         9  |   .3192203   .1471934     2.17   0.030     .0307266    .6077139
                                        10  |    .272194   .1943689     1.40   0.161     -.108762      .65315
                                            |
                          ceo_trait_A |   .0004911   .0000709     6.93   0.000     .0003522      .00063
                        ceo_control_1 |  -.0002068   .0002442    -0.85   0.397    -.0006854    .0002718
                        tmt_control_1 |   .0001256   .0002788     0.45   0.652    -.0004209     .000672
                        ceo_control_2 |  -7.14e-06   7.91e-07    -9.02   0.000    -8.69e-06   -5.59e-06
                        tmt_control_2 |   5.96e-06   1.78e-06     3.35   0.001     2.47e-06    9.45e-06
                 ceo_tmt_relationship |   .0003316   .0003541     0.94   0.349    -.0003623    .0010256
                          tmt_trait_A |   .0001271   .0000793     1.60   0.109    -.0000284    .0002826
                                      |
 c.ceo_tmt_relationship#c.tmt_trait_A |  -2.32e-06   6.91e-06    -0.34   0.737    -.0000159    .0000112
                                      |
                                _cons |  -.4752772   .1462696    -3.25   0.001    -.7619604    -.188594
-------------------------------------------------------------------------------------------------------------

------------------------------------------------------------------------------
  Random-effects parameters  |   Estimate   Std. err.     [95% conf. interval]
-----------------------------+------------------------------------------------
firm_id: Identity               |
                  var(_cons) |   .1396593   .0040585       .131927    .1478447
-----------------------------+------------------------------------------------
tmt_firm_pair_id: Identity         |
                  var(_cons) |   .0377418    .000635      .0365175    .0390072
-----------------------------+------------------------------------------------
               var(Residual) |   .1020966   .0004819      .1011564    .1030456
------------------------------------------------------------------------------
LR test vs. linear model: chi2(2) = 97786.99              Prob > chi2 = 0.0000

Note: LR test is conservative and provided only for reference.
    To reiterate, the question is this: Am I running the multilevel models correctly? The model runs and gives out result tables, but I am not sure if I’m specifying the models correctly given what I am trying to test.

    More specifically, some of the confusing I’m having revolves around issues such as: Am I using the correct grouping variables in each of the levels? Should I try to take more levels into account (e.g., industry)? Can I or should I control for the year effect in the way I’m including them in the model now?

    I am aware of the small size of the covariates for the variables of interest in the model. I think that can be taken care of by changing the scale of the variables later on.


    Any insight on how I am running the models would be much appreciated! Thank you for your time and consideration in advance.
    Tags: None
  • #2
    In a mixed model the outcome needs to be at the lowest "level" of the data hierarchy. Your outcome is measured yearly for each firm, if I understand your data correctly. Your predictor of interest is measured uniquely for the individual TMT team member on a firm by year basis, correct? If so, I do not think that your current modeling approach is correct. With yearly data on firms, you need to bring your predictor of interest to the same cadence.

    Accordingly, I think your best option is to calculate firm by year measures of the TMT members' Trait A. This could be a mean (bysort firm_id year: egen mn_tmt_trait_A = mean(tmt_trait_A)), a standard deviation (replace mean with sd in the egen statement), and/or both. This may not be ideal, but your fundamental problem is that your outcome is measured yearly for each firm. Thus, your predictors either have to be measured yearly by firm or be unique to each firm (e.g., industry).

    Comment

    • #3
      Dear Erik,

      Thank you very much for your comment. So you are suggesting that I collapse the values of main DV (TMT's Trait A) over multiple years into a single variable - Am I correct? If so, that would mean essentially I am collapsing a panel dataset into what looks like a cross-sectional data (as I would need to collapse all other yearly measured values as well). I'm not sure if I want to forego the panel data setup... do you think there would be any other way to treat yearly values? I've seen models that treat year as another level, but due to year being a bit different then firm or industry, the model was run in two steps. But I wasn't sure if such a setup is applicable to my case. Any additional help would be much appreciated! Thanks.

      https://onlinelibrary.wiley.com/doi/....1002/smj.2503
      Quigley & Graffin, 2017. Reaffirming the CEO effect is significant and much larger than chance: A comment on Fitza (2014). Strategic Management Journal, 38(3) 793-801

      Cheers,
      Chuljin

      Comment

      • #4
        Hello Chuljin,

        I want to make sure that I understand your data, and am going to focus on just four variables, as I believe these are the essential ones to the analysis:
        • firm_id
        • year
        • firm_outcome
        • tmt_trait_A
        That both firm_id and year appear in your model makes me think that, at the very least, you have yearly data on firms.

        The fact that you name your outcome is named firm_outcome is somewhat confusing. I assume that your outcome variable varies within firms

        tmt_trait_A is your key independent variable. And you said that you have this variable measured for multiple individuals per firm. I assume you also have these individual ratings measured in multiple years for each firm.

        As I said in my last post, a multilevel or mixed effects model works on the premise that the outcome variable is measured repeatedly within some grouping variable (repeatedly within firms in your case). The predictors can be measured either repeatedly within the grouping variable or be constant or enduring characteristics of the grouping variable. When you have a predictor that is measured more frequently than the outcome, then you have to think carefully about how to include that in the model. Almost without fail, you will have to do some aggregation to get it into the model.

        Your situation is that your key independent variable, tmt_trait_A, is measured for each of some number of individuals within a firm. It sounds like you have individual ratings for multiple years. But again, your outcome is not measured for each of the individuals. The outcome is something about their firm in a given year. Accordingly, you have to aggregate these individual characteristics to a firm by year cadence. This means that you would have to take the mean of all individual ratings within a firm by year. If your data was structured in a long (person-period) format, the code I gave above would do that for you.

        • 1 like

        Comment

        • #5
          I agree completely with the thrust of #4. I disagree with one detail.

          This means that you would have to take the mean of all individual ratings within a firm by year.
          Aggregating the individual ratings to the firm level doesn't necessarily entail taking the mean. While that is probably the most frequently used approach, depending on the real world meanings of and relationships among the variables, there may be good reasons to aggregate by taking some other statistic such as the median, or the minimum, or the maximum, or the most recent, or something else entirely.
          • 1 like

          Comment

          • #6
            Good point, Clyde. In my earlier post, I mentioned that one could also calculate a measure of spread for the multiple raters that could feed into the model. There is no reason one has to use the mean only.
            • 1 like

            Comment

            • #7
              Dear Erik and Clyde,

              Thank you again for your continued input. It seems I have been a bit confused about what you were proposing. So you are suggesting that due to the nature of how multilevel modeling works and how my data is structured, I need to aggregate all of the individual tmt_trait_A within a firm in a given year, to match the cadence of firm by year of the firm_outcome variable which is my DV. This would keep the panel data structure alive but essentially create a single value for tmt_trait_A for a firm in a given year.

              I think this is where my confusion or uncertainty originally came from. What I am hoping to do is to not aggregate individual values for each TMT member, but look at them separately. Let me try to elaborate a bit more.

              How my data is structured is like this:

              I have a firm-level outcome variable which is the DV. It is a type of performance rating for a firm, that is given every year and varies over time. (firm_outcome)

              Then I have the individual top management team member's Trait A values, allocated to each individual for each year as well. So for instance, if there is a CFO, COO, and CTO in a firm in my sample for years 2011, 2012, and 2013, each of CFO / COO / CTO would have separate values for each year. (tmt_trait_A)

              And the data is structured with IDs given to each firm, each individual, and each firm-individual pairs.


              Looking at the data again, I'm thinking what I want to do here is somewhat of "going up the level" in the multilevel model. In multilevel models I have seen so far, it was all going down the level (e.g., looking at individual outcomes nested within a group). I think this speaks to what you said, Erik, about IV measured more frequently than the DV.

              But the issue here is that I need to look at each individual separately. What I am trying to find out is that whether individual top management team members (who are not the CEO) will have an impact on firm outcomes over and beyond the influence of the CEO. In order to do that, I need to look at each individual separately and not as an aggregated group. This is more so because another variable of interest (for its moderation effect) is individual top management team member's relationship with the CEO. I will not be able to take that variable into account if I aggregate the top management team's info into one value per year.

              I hope I had made the situation bit clearer. Am I correct in the understanding I stated above? If so, do you have any suggestions for working around the mentioned issues? Any suggestions or comments would be greatly appreciated. Thank you!

              Comment

              • #8
                This is a helpful explanation. It sounds like you already have separate variables for Trait A values for the leadership members. And you have those on a yearly cadence. If so, you can enter them into the model as separate predictors (tmt_trait_A_CFO; tmt_trait_A_COO, tmt_trait_A_CTO). Am I missing something?
                • 1 like

                Comment

                • #9
                  Hi Erik,

                  That sounds like a good suggestion. By doing what you suggested, I would be able to put the variables at the same cadence with the firm level outcome without aggregating them all into single average. I would have to try restructuring my data to see if it would actually pan out for all firms included, but it is an approach I haven't thought about and definitely worth trying.

                  Thank you and I'll keep you posted on how it plays out! (in the meantime, any additional comments or thoughts would be welcome)

                  Best,
                  Chuljin
                  • 1 like

                  Comment

                  • #10
                    Dear Clyde Schechter Erik Ruzek

                    I also have a question concerning the application of mixed models for panel data. Specifically, my independent and dependent variables of interest are on the organizational level whereas my moderators are on the country level (e.g., stringency of environmental policy, GHG emissions etc.). I am applying a normal fixed-effect regression with cluster robust errors on the organizational level as well as year-fixed effects.

                    xtreg DV c.IV##c.MOD1 c.IV##c.MOD2 controls, fe vce(robust)

                    my panel is set by xtset organization year

                    As the fixed-effects estimator is a within estimator, I assume that I can trust the results of this regression even though my moderator variables are on the country level. It does not allow me to compare the effects of policy changes on organizations in different countries but within organizations (my panel) in general. Is my understanding correct here? If I would like to see if there are different effects for the different organizations in different countries I would have to apply mixed models?

                    Thanks for your help
                    Patrick

                    Comment

                    • #11
                      Yes, your understanding is correct. I would go a step farther and point out that "to see if there are different effects for the different organizations in different countries" you would not only need to use a mixed model, you would need the model to include random slopes.
                      • 1 like

                      Comment

                      • #12
                        Thanks Clyde Schechter for the quick response, really appreciate your help. Again for clarification - if I just want to make a general statement about the effects of my country-level moderators on my organizational IV, I can apply the standard xtreg command with fixed effects even though I have variables on different levels?

                        Comment

                        • #13
                          I am confused. You say that "my independent and dependent variables of interest are on the organizational level," however your xtset command indicates that organizations are the clusters. In a fixed effect model, cluster-level variables are omitted because they have no within-cluster variation. Perhaps you meant that your IV and DV are measured yearly and thus vary within clusters?

                          The main effect of a variable at a level higher than your cluster will be dropped from the model because, again, it does not vary with cluster in a fixed effect model. The interaction with a within-cluster predictor will be estimated, however. You cannot produce margins from this model because the main effect is dropped.

                          If you want to model the main effect in addition to the interaction and get margins, then you need to use a mixed (or random) effects model.

                          Here is some example data and code to show what I mean.
                          Code:
                          use https://www.stata-press.com/data/mlmus3/kenya
*structure - repeated measures nested w/in child nested w/in schools

xtset id rn

*fixed effects model w/ treatment assigned at the school level interacted with an observation-level variable (relyear)
xtreg ravens i.treatment##c.relyear, vce(cluster id) fe
margins treatment, at(relyear=(-.14(.05)1.8))            // not estimable
margins , at(relyear=(-.14(.05)1.8)) dydx(treatment) // not estimable 

*mixed version accounting for 3-level structure and random slope at child level
mixed ravens i.treatment##c.relyear || schoolid: || id: relyear , cov(un)
margins treatment, at(relyear=(-.14(.05)1.8))            // estimable

                          Comment

                          • #14
                            Thanks for your reply, Erik. True, my IV and DV vary within clusters (timeframe 2011-2017).

                            Code:
                            * Example generated by -dataex-. To install: ssc install dataex
clear
input double UltimateParentID int Year double ENV float DSGR_L1 double(OECD_L1 CW_L1)
4295641240 2017 5.44087665647298  0 2.9722221    5783
4295858988 2014 34.5515083440308 18 3.2777777   72.03
4295858988 2015 35.3520671834625 13 3.1111112   68.27
4295858988 2016 30.9385167110322  7 2.9444444   69.55
4295858988 2017 29.2615245214633  7 2.9444444   69.85
4295859306 2011 17.1794871794871  0 2.5555556  121.93
4295859306 2012 16.4672841094245  0 2.9444444  110.83
4295859306 2013 21.3055865229778  0 3.0555556  109.61
4295859306 2014 25.7891737891737  0 2.9444444  110.55
4295859306 2015 45.8000949667616  0 2.9444444  104.43
4295859306 2016 60.0688538188538  0 2.8333333  109.08
4295859306 2017 60.5692918192918  0 2.8888888  110.53
4295860976 2016  14.033264033264  0 2.4722221 5705.12
4295860976 2017 12.7060439560439  1 2.9722221    5783
4295865864 2011 38.0459459459459  0 4.0833335   64.94
4295865864 2012 35.4941094941094  0 4.2222223   57.51
4295865864 2013 29.7948717948717  0 3.8888888   52.39
4295865864 2014 32.4207504207504  0 4.0555553   53.79
4295865864 2015 31.5469975752389  0 4.1111112   49.59
4295865864 2016 29.9643033858768  0 4.0277777   47.57
4295865864 2017 31.8978723404255  0 3.9444444   52.01
4295866378 2011 70.4494767541602  0 3.1111112   66.72
4295866378 2012 68.7433129280081  0      3.75   19.24
4295866378 2013 68.2020149399234  0 3.6388888   13.65
4295866378 2014 71.9788673877528  0 3.6388888   14.43
4295866378 2015 73.1561302681992  0 3.6944444   10.41
4295866378 2016 71.8427660400657  0 3.8611112    6.94
4295866378 2017 69.3128493603397  0 3.8333333   70.93
4295866480 2011 68.4313725490196 68 3.1111112   66.72
4295866480 2012  89.576124567474 56      3.75   19.24
4295866480 2013 83.9965397923875 72 3.6388888   13.65
4295866480 2014 82.1940953884133 57 3.6388888   14.43
4295866480 2015 83.5994397759103 41 3.6944444   10.41
4295866480 2016 92.8566176470588 15 3.8611112    6.94
4295866480 2017 93.1421256897035  5 3.8333333   70.93
4295866790 2016 76.4606102262552  0 4.0277777  372.72
4295866790 2017 72.4421928318466  0 3.9166667  372.24
4295866983 2011                0  0 3.6111112     407
4295866983 2012 16.1906625445153  0 3.9444444  396.23
4295866983 2013 18.4599619382228  0 3.9166667  399.86
4295866983 2014 23.5836627140975  0 3.9166667  400.78
4295866983 2015 24.8447204968944  0 4.2222223  367.33
4295866983 2016 36.3739183040059  0 4.0277777  372.72
4295866983 2017 39.0338675189111  0 3.9166667  372.24
4295867015 2011 47.0530489315872  1 3.6111112     407
4295867015 2012 47.3304924806802  1 3.9444444  396.23
4295867015 2013  45.185271502035  3 3.9166667  399.86
4295867015 2014 44.4700102967594  0 3.9166667  400.78
4295867015 2015 44.0469348659003  0 4.2222223  367.33
4295867015 2016 42.1656520937096  0 4.0277777  372.72
4295867015 2017 71.9128003046615  1 3.9166667  372.24
4295867357 2011 74.2808943786813  0 3.6111112     407
4295867357 2012 72.8828410796942  1 3.9444444  396.23
4295867357 2013 77.0099893982598  2 3.9166667  399.86
4295867357 2014 72.9262709769089  0 3.9166667  400.78
4295867357 2015 77.0162835249042  0 4.2222223  367.33
4295867357 2016 78.4504030074149  1 4.0277777  372.72
4295867357 2017 76.0641370718923  2 3.9166667  372.24
4295867377 2011  33.397190293742  0 3.6111112     407
4295867377 2012 63.3349084314576  0 3.9444444  396.23
4295867377 2013 64.2050370451609  0 3.9166667  399.86
4295867377 2014 63.0012041575122  0 3.9166667  400.78
4295867377 2015 64.1810344827586  0 4.2222223  367.33
4295867377 2016 40.8958986656828  0 4.0277777  372.72
4295867377 2017 41.5608903605592  0 3.9166667  372.24
4295868093 2017 36.8995789387621  0 3.9166667  372.24
4295868112 2011 14.1975308641975  0 3.6111112     407
4295868112 2012 19.3672839506172  1 3.9444444  396.23
4295868112 2013  18.287037037037  1 3.9166667  399.86
4295868112 2014 23.5339506172839  1 3.9166667  400.78
4295868112 2015 21.2962962962962  1 4.2222223  367.33
4295868112 2016 18.8271604938271  0 4.0277777  372.72
4295868112 2017 64.8119908859713  0 3.9166667  372.24
4295868416 2011 81.9483150252156  3 3.6111112     407
4295868416 2012 83.4472713575705  1 3.9444444  396.23
4295868416 2013 85.4223393353828  5 3.9166667  399.86
4295868416 2014 85.6642958392693  1 3.9166667  400.78
4295868416 2015 85.0033675072962  3 4.2222223  367.33
4295868416 2016 82.9752886714537  2 4.0277777  372.72
4295868416 2017 87.0392298122048  0 3.9166667  372.24
4295869055 2017 37.0058896817306  0 3.0833333  836.97
4295869278 2011 2.33079336620349  0 3.0833333  881.31
4295869278 2012 1.58102766798419  0 3.1666667  840.75
4295869278 2013 1.58102766798419  0 3.0555556  853.95
4295869278 2014 1.75376957985653  0 3.2222223  870.32
4295869278 2015 2.32919254658385  0 3.1111112  829.47
4295869278 2016 14.1481890633631  0 3.0277777  834.14
4295869278 2017 35.8103858347575  0 3.0833333  836.97
4295869822 2011                0  0 3.0833333  881.31
4295869822 2012                0  0 3.1666667  840.75
4295869822 2013                0  0 3.0555556  853.95
4295869822 2014                0  0 3.2222223  870.32
4295869822 2015                0  1 3.1111112  829.47
4295869822 2016 4.67625899280575  0 3.0277777  834.14
4295869822 2017 10.9271523178807  0 3.0833333  836.97
4295870063 2011 61.1204430332945  9 3.0833333  881.31
4295870063 2012 51.2465277777777 10 3.1666667  840.75
4295870063 2013 58.1161748220823 16 3.0555556  853.95
4295870063 2014 62.8256345241261 21 3.2222223  870.32
4295870063 2015 72.3091498671667 18 3.1111112  829.47
end
format %ty Year
                            Code:
                            xtset UltimateParentID Year

xtreg ENV c.DSGR_L1##c.OECD_L1 c.DSGR_L1##c.CW_L1 i.Year, fe vce(robust)
                            ENV = organizational level
                            DSGR_L1 = organizational level
                            OECD_L1 = country level
                            CW_L1 = country level

                            Comment

                            • #15
                              This all looks right. You get the interaction coefficients but they are hard to interpret because you don't have the main effect. Plenty of folks are fine with that and if that is the norm in your field, go with it.

                              Comment

                              Previous 1 2 Next
                              Working...
                              X