high coeffients in simple OLS and even spline

[Sujata Luiss]

I am running a very simple OLSregeession with cross sectional datt having 50companies and 4628 directors' data.My coeffients on one of my intercation erm is comin quite high like 22. The interacion is between a dummy variable and a continuous variable but the continuoys varibale is having many 0 vales in the observatiom. I don't know if such high coeffients is acceptable or not. I am new to stata and so may ne I have asked a very simple qquestion. But any help would be highly appreciated.

[Sujata Luiss]

sorry that will be 500 companies.

[Carlo Lazzaro]

Sujata:
welcome to the list.
Please note that your chances of getting helpful replies are conditional on posting what you typed and what Stata gave you back (as per FAQ). Thanks.

[Sujata Luiss]

Thank you. I am posting my code here. the frst term is an nteraction term whoch is showing me quite high coefficients

Source | SS df MS Number of obs = 321
-------------+---------------------------------- F(57, 263) = 3.99
Model | 2878.28169 57 50.49617 Prob > F = 0.0000
Residual | 3331.36583 263 12.6667902 R-squared = 0.4635
-------------+---------------------------------- Adj R-squared = 0.3472
Total | 6209.64752 320 19.4051485 Root MSE = 3.559

----------------------------------------------------------------------------------------
pb | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-----------------------+----------------------------------------------------------------
int_2 | 22.33806 11.96909 1.87 0.063 -1.229379 45.90549
centered_board_ind_two | -9.565016 7.35646 -1.30 0.195 -24.05007 4.920039
prom_ceo | -1.066417 .579557 -1.84 0.067 -2.20758 .0747448
ceo_dual | -.7370154 .5392245 -1.37 0.173 -1.798762 .3247312
prom_dir_bd | 1.259703 1.546017 0.81 0.416 -1.784443 4.303849
business_grp_dummy | -.5866 .5135335 -1.14 0.254 -1.59776 .4245603
ceo_tenure_2013 | -.0001641 .024233 -0.01 0.995 -.0478795 .0475513
ind_prom_shares | -2.591756 1.023855 -2.53 0.012 -4.607752 -.57576
lmkt_cpt_prev | 1.320849 .1789101 7.38 0.000 .9685703 1.673127

[Sujata Luiss]

the interaction term int_2 is a combination of one dummy variable and another continuous variable centered_board_ind_two. But this continuous variable is having many '0' values in the observation and I need those 0 values also. The way the coefficient came left me a bit shock d as I am not sure what can be the reason or whether it is good to be it like this.I saw some paper in journal of management studies having quite high coefficient estimates. So I am not sure what to do,
.

[Nick Cox]

Here are the coefficients a little more readably.

  |                  int_2    22.33806 |
  | centered_board_ind_two   -9.565016 |
  |               prom_ceo   -1.066417 |
  |               ceo_dual   -.7370154 |
  |            prom_dir_bd    1.259703 |
  |     business_grp_dummy      -.5866 |
  |        ceo_tenure_2013   -.0001641 |
  |        ind_prom_shares   -2.591756 |
  |          lmkt_cpt_prev    1.320849 |

Such coefficients are slopes or gradients and can't be interpreted without knowing the units and spread of variables. So, what can you tell us about your response (outcome, dependent variable)? Can you show us the results of

summarize pb, detail

[Sujata Luiss]

do you want me to show the detail for each variable?
I can do that...my outcome variable is pb (price to book ratio)

[Sujata Luiss]

. summarize pb, detail

PB
-------------------------------------------------------------
Percentiles Smallest
1% .218997 .173974
5% .403228 .188193
10% .513983 .192881 Obs 493
25% .892481 .204482 Sum of Wgt. 493

50% 1.761988 Mean 3.059937
Largest Std. Dev. 4.01664
75% 3.512082 25.0471
90% 6.668971 28.59596 Variance 16.1334
95% 9.976144 31.11458 Skewness 4.182878
99% 24.74946 39.9717 Kurtosis 28.40461

. summarize int_2, detail

int_2
-------------------------------------------------------------
Percentiles Smallest
1% -.0094224 -.0094224
5% -.0094224 -.0094224
10% -.0094224 -.0094224 Obs 489
25% -.0094224 -.0094224 Sum of Wgt. 489

50% 0 Mean -.0002017
Largest Std. Dev. .0221964
75% 0 .1334347
90% 0 .1444237 Variance .0004927
95% 0 .1572442 Skewness 6.38427
99% .1334347 .2405775 Kurtosis 50.09779

. summarize centered_board_ind_two, detail

centered_board_ind_two
-------------------------------------------------------------
Percentiles Smallest
1% -.0094224 -.0094224
5% -.0094224 -.0094224
10% -.0094224 -.0094224 Obs 492
25% -.0094224 -.0094224 Sum of Wgt. 492

50% -.0094224 Mean 1.17e-10
Largest Std. Dev. .0345073
75% -.0094224 .1723957
90% -.0094224 .1905776 Variance .0011908
95% .0905776 .2127998 Skewness 3.882912
99% .1572442 .2405775 Kurtosis 18.4222

.

[Sujata Luiss]

I am sending this for my outcome variable, independent and interaction variables.....

[Phil Bromiley]

Sujata - as you may have seen, your output is very hard to read. If you're going to paste such results, you'll often find a fixed space font like Courier New makes things readable. Also, look at the factor variable notation - it makes handling interactions much easier.

If you want to look into this, do predictive margins for the range of your x variables and see if the differences in predictions make sense. While alternatively, with a linear model, you can just multiple the coefficient times the variable's standard error to give you an estimate of the chance in predicted value for a one standard deviation change in the x, this might be a little harder than the margins approach. As Nick pointed out, you have to do this calculation (or the margins equivalent) to say whether the effect sizes make sense.