Hi,
I am working on a paper concerning different strategies in sourcing Intermediate inputs in production. My outcome variable is a categorical variable taking values 1-5. I want to check whether there are any significant differences in the organization of the firm across these categories. To do that I have been running -mlogit-, with some independent variables. However, since these are European firms, I hypothesise that there might be differences between Northern and Southern Europe, and I would like to see if there are any differences in the structure of the firm also across these two regions. I was wondering if this can be done by just splitting the sample in North and South, and then run two independent -mlogit x1 x2 ... if North == 1- and -mlogit x1 x2 ... if South == 1- and then just list them next to each other, and compare like (Note: it is not a causal study, I am just trying to see whether there are significant differences in the structure of the firm):
For example, with these results I could say something like: Both in the South and the North being an exporter increases the relative probability that the firm is using FO, DIFO, or FI over DO.
In the South, exporting significantly increases the log odds of using DI over DO, while there is no significant effect in the North. This suggests that exporting and foreign sourcing is positively related in the North, while in the South, the relative probability of all other sourcing strategies compared to DO increases if a firm also exports some of its production.
I guess I could need some advice on how to formulate an interpretation that is not sounding causal, since this is not a causal study. It should not sound like exporting causes the relative probability of FI to increase, but rather that firms that export are relatively more likely to use FI than DO. :D
All firms in the sample is either in the North or in the South, so I was wondering if this could possibly be done within the same model using interaction terms, but I was not certain of how to interpret the results then. I tried making an example, but it just does not make sense to me:
Here exportern is an interaction term: exporting*North.
I am working on a paper concerning different strategies in sourcing Intermediate inputs in production. My outcome variable is a categorical variable taking values 1-5. I want to check whether there are any significant differences in the organization of the firm across these categories. To do that I have been running -mlogit-, with some independent variables. However, since these are European firms, I hypothesise that there might be differences between Northern and Southern Europe, and I would like to see if there are any differences in the structure of the firm also across these two regions. I was wondering if this can be done by just splitting the sample in North and South, and then run two independent -mlogit x1 x2 ... if North == 1- and -mlogit x1 x2 ... if South == 1- and then just list them next to each other, and compare like (Note: it is not a causal study, I am just trying to see whether there are significant differences in the structure of the firm):
Code:
. mlogit sourcingmode tfp2008 exporter if coresample == 1 & north == 1, robust base(1)
Iteration 0: log pseudolikelihood = -3185.0168
Iteration 1: log pseudolikelihood = -3092.6962
Iteration 2: log pseudolikelihood = -3089.8223
Iteration 3: log pseudolikelihood = -3089.7949
Iteration 4: log pseudolikelihood = -3089.7949
Multinomial logistic regression Number of obs = 2,261
Wald chi2(8) = 162.29
Prob > chi2 = 0.0000
Log pseudolikelihood = -3089.7949 Pseudo R2 = 0.0299
------------------------------------------------------------------------------
| Robust
sourcingmode | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
DO | (base outcome)
-------------+----------------------------------------------------------------
DI |
tfp2008 | .0463079 .1671139 0.28 0.782 -.2812293 .3738451
exporter | -.165352 .1630326 -1.01 0.310 -.4848899 .154186
_cons | -1.075222 .1230861 -8.74 0.000 -1.316466 -.8339775
-------------+----------------------------------------------------------------
FO |
tfp2008 | .2543847 .0928671 2.74 0.006 .0723686 .4364009
exporter | 1.012922 .1155234 8.77 0.000 .7865008 1.239344
_cons | -.4321649 .0975584 -4.43 0.000 -.6233759 -.2409539
-------------+----------------------------------------------------------------
DIFO |
tfp2008 | .2472427 .1828191 1.35 0.176 -.1110762 .6055615
exporter | .8684782 .1846155 4.70 0.000 .5066384 1.230318
_cons | -1.729958 .1585914 -10.91 0.000 -2.040791 -1.419125
-------------+----------------------------------------------------------------
FI |
tfp2008 | .552963 .1579696 3.50 0.000 .2433483 .8625778
exporter | 2.014189 .2645019 7.62 0.000 1.495775 2.532603
_cons | -2.737447 .2499555 -10.95 0.000 -3.227351 -2.247544
------------------------------------------------------------------------------
Code:
. mlogit sourcingmode tfp2008 exporter if coresample == 1 & south == 1, robust base(1)
Iteration 0: log pseudolikelihood = -5227.2017
Iteration 1: log pseudolikelihood = -4944.6407
Iteration 2: log pseudolikelihood = -4935.4004
Iteration 3: log pseudolikelihood = -4935.0236
Iteration 4: log pseudolikelihood = -4935.0224
Iteration 5: log pseudolikelihood = -4935.0224
Multinomial logistic regression Number of obs = 4,336
Wald chi2(8) = 438.85
Prob > chi2 = 0.0000
Log pseudolikelihood = -4935.0224 Pseudo R2 = 0.0559
------------------------------------------------------------------------------
| Robust
sourcingmode | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
DO | (base outcome)
-------------+----------------------------------------------------------------
DI |
tfp2008 | .8361852 .1292515 6.47 0.000 .582857 1.089513
exporter | .309192 .12668 2.44 0.015 .0609037 .5574802
_cons | -1.939009 .1045761 -18.54 0.000 -2.143974 -1.734043
-------------+----------------------------------------------------------------
FO |
tfp2008 | .6267666 .079925 7.84 0.000 .4701166 .7834167
exporter | 1.34326 .0871128 15.42 0.000 1.172522 1.513998
_cons | -1.285744 .0778412 -16.52 0.000 -1.43831 -1.133178
-------------+----------------------------------------------------------------
DIFO |
tfp2008 | 1.275918 .1410801 9.04 0.000 .9994066 1.55243
exporter | 1.565335 .1866388 8.39 0.000 1.19953 1.93114
_cons | -3.041633 .1732831 -17.55 0.000 -3.381261 -2.702004
-------------+----------------------------------------------------------------
FI |
tfp2008 | 1.39548 .2435528 5.73 0.000 .9181253 1.872835
exporter | 2.857254 .4582993 6.23 0.000 1.959003 3.755504
_cons | -4.988943 .4496186 -11.10 0.000 -5.87018 -4.107707
------------------------------------------------------------------------------
In the South, exporting significantly increases the log odds of using DI over DO, while there is no significant effect in the North. This suggests that exporting and foreign sourcing is positively related in the North, while in the South, the relative probability of all other sourcing strategies compared to DO increases if a firm also exports some of its production.
I guess I could need some advice on how to formulate an interpretation that is not sounding causal, since this is not a causal study. It should not sound like exporting causes the relative probability of FI to increase, but rather that firms that export are relatively more likely to use FI than DO. :D
All firms in the sample is either in the North or in the South, so I was wondering if this could possibly be done within the same model using interaction terms, but I was not certain of how to interpret the results then. I tried making an example, but it just does not make sense to me:
Code:
. mlogit sourcingmode tfp2008 exporter exportern if coresample == 1, robust base(1)
Iteration 0: log pseudolikelihood = -8560.3337
Iteration 1: log pseudolikelihood = -8128.4708
Iteration 2: log pseudolikelihood = -8099.7006
Iteration 3: log pseudolikelihood = -8099.396
Iteration 4: log pseudolikelihood = -8099.3953
Multinomial logistic regression Number of obs = 6,597
Wald chi2(12) = 714.31
Prob > chi2 = 0.0000
Log pseudolikelihood = -8099.3953 Pseudo R2 = 0.0538
------------------------------------------------------------------------------
| Robust
sourcingmode | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
DO | (base outcome)
-------------+----------------------------------------------------------------
DI |
tfp2008 | .5797641 .1048356 5.53 0.000 .37429 .7852382
exporter | -.0103951 .1059574 -0.10 0.922 -.2180678 .1972776
exportern | .4099698 .1304038 3.14 0.002 .1543831 .6655566
_cons | -1.649886 .078551 -21.00 0.000 -1.803843 -1.495929
-------------+----------------------------------------------------------------
FO |
tfp2008 | .5399869 .0619008 8.72 0.000 .4186636 .6613101
exporter | 1.036042 .0716425 14.46 0.000 .8956254 1.176459
exportern | .5510883 .0742808 7.42 0.000 .4055007 .696676
_cons | -.9994194 .0597341 -16.73 0.000 -1.116496 -.8823427
-------------+----------------------------------------------------------------
DIFO |
tfp2008 | .8820604 .1147819 7.68 0.000 .657092 1.107029
exporter | 1.029363 .1329755 7.74 0.000 .7687362 1.289991
exportern | .5856167 .117866 4.97 0.000 .3546036 .8166298
_cons | -2.51832 .1146602 -21.96 0.000 -2.74305 -2.29359
-------------+----------------------------------------------------------------
FI |
tfp2008 | 1.005622 .136158 7.39 0.000 .7387571 1.272487
exporter | 1.694092 .2332873 7.26 0.000 1.236858 2.151327
exportern | 1.387403 .1277538 10.86 0.000 1.13701 1.637796
_cons | -3.83035 .2153664 -17.79 0.000 -4.25246 -3.408239
------------------------------------------------------------------------------
