Economics of Crime Dissertation - terrorist attacks and hate crime in London

[Emily Nicholas]

Hi Everyone,

I am new to Statalist so please let me know if I am not using the forum correctly.

For my dissertation, I am investigating how Islamist terror attacks in London have affected Islamophobic hate crime between Jan2015-Dec2018.

I have obtained a panel dataset containing all recorded incidents of hate crime, disaggregated by:
- Week
- Type of hate crime (Islamophobic, Homophobic, Anti-Semetic...)

My goal is to:
1) Identify the contemporaneous increase in Islamophobic hate crime attributable to UK and non-UK terror attack
2) Using leads and lags, determine how long the attack has a significant effect on hate crime

I am employing a difference in difference specification whereby Islamophobic hate crime is my treatment group and Homophobic&Transgender hate crime is my control. Within the time period specified 8 non-UK terror attacks occur and 3 UK terror attacks - I am estimating their effects separately. Therefore, instead of having a normal pre and post time dummy, I have a non-UK attack dummy and a UK attack dummy that only takes the value of 1 if an attack occured in that week and 0 if not. The interaction of 6 leads and 12 lags of the terror attack with the treatment indicator (islamophobic hate crime) then shows the dynamic effects of the attack (like an event study).


I would like to know:

1) How to decide the number of lags to use for both terror_attack_uk and terror_attack_nonuk - as it appears I cannot use AIC or BIC with factor variables

2) Why when I use a log transformation of hate_crime do my coefficients fall into insignificance that were previously significant in levels (eg. islam#L4. terror_a~_uk)

3) Does anyone have any advice on what I should do about my standard errors as I understand there as some difficulties with clustering errors when the group number is small (in my case only 2)

When I type:

    
    reg hate_crime i.islam##L(-6/12)terror_attack_uk i.islam##L(-6/12)terror_attack_nonuk i.months
    estimates store lags

    reg log_hate_crime i.islam##L(-6/12)terror_attack_uk i.islam##L(-6/12)terror_attack_nonuk i.months
    estimates store lags_logs 
    
    estimates table lags lags_logs, star stats (N r2 r2_a)

I get:

----------------------------------------------
    Variable |     lags          lags_logs    
-------------+--------------------------------
       islam |
          1  | -21.377506***    -.6725765***  
             |
         F6. |
terror_a~_uk |
          1  | -1.1976008        .0169865     
             |
         F5. |
terror_a~_uk |
          1  | -6.0323428      -.13274829     
             |
         F4. |
terror_a~_uk |
          1  |  -2.165258      -.04263706     
             |
         F3. |
terror_a~_uk |
          1  | -6.8671289      -.13549132     
             |
         F2. |
terror_a~_uk |
          1  |  1.8670848       .03947542     
             |
          F. |
terror_a~_uk |
          1  |  7.1005283       .15395964     
             |
terror_a~_uk |
          1  | -2.1005725      -.02171693     
             |
          L. |
terror_a~_uk |
          1  |  .06818554       .00485655     
             |
         L2. |
terror_a~_uk |
          1  | -19.634773      -.47634242     
             |
         L3. |
terror_a~_uk |
          1  |  12.035843       .32351044     
             |
         L4. |
terror_a~_uk |
          1  | -8.6024303      -.25339895     
             |
         L5. |
terror_a~_uk |
          1  |     9.0035       .25179803     
             |
         L6. |
terror_a~_uk |
          1  | -7.8475323      -.17354257     
             |
         L7. |
terror_a~_uk |
          1  |  26.694596       .51895765     
             |
         L8. |
terror_a~_uk |
          1  | -22.446848      -.38081646     
             |
         L9. |
terror_a~_uk |
          1  |  23.416006       .45233668     
             |
        L10. |
terror_a~_uk |
          1  | -12.317935      -.10430467     
             |
        L11. |
terror_a~_uk |
          1  |  23.638811       .40975008     
             |
        L12. |
terror_a~_uk |
          1  |  1.1208767       .15356868     
             |
       islam#|
         F6. |
terror_a~_uk |
        1 1  |  -8.676533      -.20006207     
             |
       islam#|
         F5. |
terror_a~_uk |
        1 1  |  18.108077        .2217817     
             |
       islam#|
         F4. |
terror_a~_uk |
        1 1  |  -18.78461      -.18068171     
             |
       islam#|
         F3. |
terror_a~_uk |
        1 1  |  27.216154       .17708343     
             |
       islam#|
         F2. |
terror_a~_uk |
        1 1  | -38.892687      -.57497117     
             |
       islam#|
          F. |
terror_a~_uk |
        1 1  |  30.324231       .25696793     
             |
       islam#|
terror_a~_uk |
        1 1  | -5.0007639       .48743158     
             |
       islam#|
          L. |
terror_a~_uk |
        1 1  |  36.432308        .3646363     
             |
       islam#|
         L2. |
terror_a~_uk |
        1 1  |  72.891159*      1.5302475     
             |
       islam#|
         L3. |
terror_a~_uk |
        1 1  | -24.459615      -.38034502     
             |
       islam#|
         L4. |
terror_a~_uk |
        1 1  |  81.783082**     1.4812173     
             |
       islam#|
         L5. |
terror_a~_uk |
        1 1  | -43.351538      -.51331752     
             |
       islam#|
         L6. |
terror_a~_uk |
        1 1  |   86.82788**      1.611772     
             |
       islam#|
         L7. |
terror_a~_uk |
        1 1  | -70.624934*     -.97402854     
             |
       islam#|
         L8. |
terror_a~_uk |
        1 1  |  65.324231*      1.1055779     
             |
       islam#|
         L9. |
terror_a~_uk |
        1 1  | -49.259385      -.62688005     
             |
       islam#|
        L10. |
terror_a~_uk |
        1 1  |  36.994539       .57201323     
             |
       islam#|
        L11. |
terror_a~_uk |
        1 1  | -36.529692      -.58465277     
             |
       islam#|
        L12. |
terror_a~_uk |
        1 1  |  .46484618      -.16912405     
             |
         F6. |
terror_a~nuk |
          1  | -7.1493615      -.16278855     
             |
         F5. |
terror_a~nuk |
          1  | -4.8728007      -.08746172     
             |
         F4. |
terror_a~nuk |
          1  | -7.2411421       -.1567165     
             |
         F3. |
terror_a~nuk |
          1  |  -4.994894      -.11584777     
             |
         F2. |
terror_a~nuk |
          1  | -12.091465**    -.32493462*    
             |
          F. |
terror_a~nuk |
          1  | -1.6816103      -.03685018     
             |
terror_a~nuk |
          1  | -4.0880788       -.0872098     
             |
          L. |
terror_a~nuk |
          1  | -8.4816103       -.2391452     
             |
         L2. |
terror_a~nuk |
          1  | -2.0019312       -.0183154     
             |
         L3. |
terror_a~nuk |
          1  | -9.0982289*     -.20351647     
             |
         L4. |
terror_a~nuk |
          1  | -3.9138039      -.05984587     
             |
         L5. |
terror_a~nuk |
          1  | -3.4138039      -.03410763     
             |
         L6. |
terror_a~nuk |
          1  | -2.9367106      -.02984166     
             |
         L7. |
terror_a~nuk |
          1  | -6.4039855      -.14022711     
             |
         L8. |
terror_a~nuk |
          1  | -5.2564695      -.08723422     
             |
         L9. |
terror_a~nuk |
          1  | -5.7564695       -.1153905     
             |
        L10. |
terror_a~nuk |
          1  | -5.0649093      -.08016198     
             |
        L11. |
terror_a~nuk |
          1  | -2.0054019      -.04993284     
             |
        L12. |
terror_a~nuk |
          1  | -4.8276962      -.12672174     
             |
       islam#|
         F6. |
terror_a~nuk |
        1 1  |  3.8471256      -.00753235     
             |
       islam#|
         F5. |
terror_a~nuk |
        1 1  |  .22859812      -.03079796     
             |
       islam#|
         F4. |
terror_a~nuk |
        1 1  |  4.6241698       .06673647     
             |
       islam#|
         F3. |
terror_a~nuk |
        1 1  | -1.6333022      -.03209505     
             |
       islam#|
         F2. |
terror_a~nuk |
        1 1  |  7.5883132       .15263939     
             |
       islam#|
          F. |
terror_a~nuk |
        1 1  | -3.0333022      -.04284958     
             |
       islam#|
terror_a~nuk |
        1 1  |  6.3883132       .19476986     
             |
       islam#|
          L. |
terror_a~nuk |
        1 1  |  11.966698       .37683497     
             |
       islam#|
         L2. |
terror_a~nuk |
        1 1  |  6.5441722       .23944214     
             |
       islam#|
         L3. |
terror_a~nuk |
        1 1  |  10.377506       .26130306     
             |
       islam#|
         L4. |
terror_a~nuk |
        1 1  |  4.7108388      -.02783756     
             |
       islam#|
         L5. |
terror_a~nuk |
        1 1  |  9.5441722       .31744718     
             |
       islam#|
         L6. |
terror_a~nuk |
        1 1  |  3.7108388       .08624939     
             |
       islam#|
         L7. |
terror_a~nuk |
        1 1  |  7.2108388       .21742864     
             |
       islam#|
         L8. |
terror_a~nuk |
        1 1  |  4.7108388        .0616424     
             |
       islam#|
         L9. |
terror_a~nuk |
        1 1  |  5.7108388       .11297066     
             |
       islam#|
        L10. |
terror_a~nuk |
        1 1  |  4.2363179       .04694795     
             |
       islam#|
        L11. |
terror_a~nuk |
        1 1  | -2.1605942      -.10295594     
             |
       islam#|
        L12. |
terror_a~nuk |
        1 1  | -1.4259473      -.11681151     
             |
      months |
          2  |  1.6129858        .0184758     
          3  |  8.1700475**     .27414883**   
          4  |  4.1139291       .11483988     
          5  |  6.9163725*      .20252438*    
          6  |  14.482916***    .38016868***  
          7  |  12.909722***    .31166969***  
          8  |  5.2447191*      .10997252     
          9  |  5.4875089*      .19369931*    
         10  |  6.6781727**     .22161665**   
         11  |  6.2344362*      .20912893*    
         12  |  2.1278858       .06468782     
             |
       _cons |  39.599843***    3.6107683***  
-------------+--------------------------------
           N |        380             380     
          r2 |  .70060705       .65530434     
        r2_a |  .61006897       .55106648     
----------------------------------------------
      legend: * p<0.05; ** p<0.01; *** p<0.001

Thank you very much for any help!

Emily

[Phil Bromiley]

Welcome to state a list. You did not get a quick answer. You will increase the chances of a useful answer by following the FAQ on asking questions – provide Stata code in code delimiters, readable Stata output, and sample data using dataex. You should also cut back your post to the minimum necessary to illustrate your problem. This long long long list of variables just makes it harder for the reader.

1) How to decide the number of lags to use for both terror_attack_uk and terror_attack_nonuk - as it appears I cannot use AIC or BIC with factor variables

You can use AIC and BIC, you simply have to run the model separately. And maybe you can't use that format for setting up your lags – I don't know..

2) Why when I use a log transformation of hate_crime do my coefficients fall into insignificance that were previously significant in levels (eg. islam#L4. terror_a~_uk)

Whenever you change your model, you can change your results. So a log transformation clearly changes the dependent variable. I might worry that your results are dependent on a very few extreme values for the dependent variable.
You have a count dependent variable. It is quite likely you should be using Poisson or negative binomial estimators. You also may need to look at zero inflated estimators that you probably have a lot of zeros..

3) Does anyone have any advice on what I should do about my standard errors as I understand there as some difficulties with clustering errors when the group number is small (in my case only 2)
Your example doesn't cluster standard errors. I'm not sure what this two cases is. If this is panel data, you may need to use a panel estimator – look at the XT documentation. Before I worried about whether the two observations per group is too small, I would see if you get the same standard errors with and without clustering.

[Sven-Kristjan Bormann]

Before worrying about the right estimation approach, I would simply plot the data first. Then you can probably already see which lags and leads are useful to employ.
Besides that, I would run additionally a logistic, logit or probit estimation to estimate the impact of leads and lags of terror attack on the probability that an Islamophobic hate crime occurs.
You might have to add interactions between the categories of hate crimes because the persons who commit hate crimes, might commit several types of hate crimes and just switch "targets" before and after a terror attack.
Therefore, your Diff-in-Diff approach might not be correct, because the categories of hate crimes are not necessarily exclusive.