R help - Oct 2011 - Coefficients for lagged plm model variables not calculated

Hello,

So I am afraid I am having a recurring problem that I just can't figure out.
I am using the plm package to conduct a panel analysis - although I am not
sure if the problem is arising as a result of the plm package or something
more general.

I am trying to run a fixed effects model with effects over time and
individual. The model has various lags, and the problem is that these lags
do not seem to always be taken into account when the plm model is calculated
(i.e no coefficients are calculated for some of the lagged regressors
when
>summary(plm.object) is called).
I have used exactly the same ".txt" file before, with exactly the same
code,
bar the name of the model objects, successfully. If I use <lm> instead of
<plm> it also will usually take into account the lagged regressors when
producing a &lt;summary(model.object).

The code I am using is:

&gt; b<-plm.data(b,index=c("E","M"))
>
b.fetw<-plm(B~lag(B,k=1)+Ma+lag(Ma,k=1)+Pa+lag(Pa,k=1)+Ya+lag(Ya,k=1)+F+lag(F,k=1)+CS+R+lag(R,k=1)+G+I,
> data=b, model="within", effect="twoways")
> summary(b.fetw)
Twoways effects Within Model

Call:
plm(formula = B ~ lag(B, k = 1) + Ma + lag(Ma, k = 1) + Pa + 
    lag(Pa, k = 1) + Ya + lag(Ya, k = 1) + F + lag(F, k = 1) + 
    CS + R + lag(R, k = 1) + G + I, data = b, effect = "twoways", 
    model = "within")

Unbalanced Panel: n=7, T=146-152, N=1041

Residuals :
    Min.  1st Qu.   Median     Mean  3rd Qu.     Max. 
-7150.00  -724.00   -73.00    -1.29   626.00 18100.00 

Coefficients :
                  Estimate  Std. Error t-value  Pr(>|t|)    
lag(B, k = 1)   3.7260e-01  3.3224e-02 11.2145 < 2.2e-16 ***
Ma              1.9957e+12  1.7898e+13  0.1115  0.911242    
lag(Ma, k = 1) -2.5335e+00  8.7506e-01 -2.8953  0.003883 ** 
Pa              9.8872e+11  4.2581e+13  0.0232  0.981480    
Ya             -4.0497e+12  4.8527e+13 -0.0835  0.933511    
F              -3.9471e+12  1.3589e+13 -0.2905  0.771531    
CS             -4.5666e+09  1.3941e+10 -0.3276  0.743321    
R               1.8369e+13  3.8520e+14  0.0477  0.961976    
G              -8.0416e+13  1.8930e+14 -0.4248  0.671086    
I               2.9025e+15  9.8887e+15  0.2935  0.769195    
---
Signif. codes:  0 ?***? 0.001 ?**? 0.01 ?*? 0.05 ?.? 0.1 ? ? 1 

Total Sum of Squares:    3210200000
Residual Sum of Squares: 2794400000
F-statistic: 12.9582 on 10 and 871 DF, p-value: < 2.22e-16

#Now the model seems to have been read correctly, but there are no
coefficients for lags of "Pa", "Ya", "F" or
"R"

I considered whether I need to convert all my data into <ts> data and then
compile it with ts.union to create a time series data frame in case there
was a problem with using the <lag> operator on my data frame, but my panel
entities are not time series data so this approach seemed flawed.

I have also used dynformula but this also seemed to lead to the same
outcome.

If on the other hand I used a simple linear model, there was no problem.
Example:
>
b.fetw<-lm(B~lag(B,k=1)+Ma+lag(Ma,k=1)+Pa+lag(Pa,k=1)+Ya+lag(Ya,k=1)+F+lag(F,k=1)+CS+R+lag(R,k=1)+G+I,data=b)
> summary(b.fetw)
Call:
lm(formula = B ~ lag(B, k = 1) + Ma + lag(Ma, k = 1) + Pa + lag(Pa, 
    k = 1) + Ya + lag(Ya, k = 1) + F + lag(F, k = 1) + CS + R + 
    lag(R, k = 1) + G + I, data = b)

Residuals:
       Min         1Q     Median         3Q        Max 
-8.802e-12 -6.201e-13 -2.103e-13  1.976e-13  2.682e-10 

Coefficients: (5 not defined because of singularities)
                 Estimate Std. Error    t value Pr(>|t|)    
(Intercept)    -2.734e-11  1.715e-12 -1.595e+01   <2e-16 ***
lag(B, k = 1)   1.000e+00  5.859e-17  1.707e+16   <2e-16 ***
Ma             -1.274e-16  2.416e-16 -5.270e-01    0.598    
lag(Ma, k = 1)         NA         NA         NA       NA    
Pa             -3.583e-16  3.419e-16 -1.048e+00    0.295    
lag(Pa, k = 1)         NA         NA         NA       NA    
Ya              4.492e-16  5.023e-16  8.940e-01    0.371    
lag(Ya, k = 1)         NA         NA         NA       NA    
F               2.251e-16  1.072e-16  2.099e+00    0.036 *  
lag(F, k = 1)          NA         NA         NA       NA    
CS             -2.294e-20  1.378e-19 -1.660e-01    0.868    
R              -1.001e-15  2.933e-15 -3.410e-01    0.733    
lag(R, k = 1)          NA         NA         NA       NA    
G               1.763e-16  1.561e-15  1.130e-01    0.910    
I               1.677e-14  8.304e-14  2.020e-01    0.840    
---
Signif. codes:  0 ?***? 0.001 ?**? 0.01 ?*? 0.05 ?.? 0.1 ? ? 1 


It was only this afternoon, that I swear I was using the same code, and the
same .txt file and having no problem!! I'm sure the answer is right in front
of me, but I have run out of ideas.

If anyone had any thoughts that may shed some light on this I would be most
grateful.

An example of an extract from my data is below.
There are sporadic NA values in the Q and B columns. All other columns are
filled.
M is my time variable
E are my entities (Q1 - Q2)
Sorry if it looks a bit messy in copy /paste!


M	E	Q	B	Ma	Ya	Pa	F	CS	G	R	I
12	Q1	1.038461538	13915.3466	2280	2760	1593	13885	0	0	5.1	2.5
13	Q1	0.88	15871.51088	2194	2711	1279	17658	0	-16.76	100.9	2.5
14	Q1	1.333333333	16374.10791	2344	2857	1373	14469	0	-26.88	49	2.5
15	Q1	0.545454545	15907.40383	2770	3029	1194	13142	0	-56.17	53.8	2.5
16	Q1	0.95	17681.04616	3242	2802	1601	14545	0	-19.48	120	2.5
17	Q1	1.136363636	13422.43498	4444	3456	1811	16460	0	-30.76	286.5	2.5
18	Q1	1.384615385	15766.59589	4642	3090	2164	14604	0	-20.66	129.9	2.5
21	Q1	1.409090909	14014.20695	3458	2414	1453	12694	2920873	-65.79	21.6	
167	Q2	3.777777778	11630.12367	2153	3229	4196	12605	0	-93.83	155.1	-1.25
168	Q2	4.352941176	11683.55864	1774	3192	3374	9886	0	-79.73	136.6	-1.25
171	Q2	NA	NA	1394	1630	1948	9083	5308514	102.92	115.6	-1.25
172	Q2	5.315789474	13324.71911	1308	1762	1830	10341	0	-3.54	3.7	1.25
173	Q2	3.105263158	12577.84126	1313	1809	1562	9132	0	-15.99	80.2	1.25
174	Q2	1	15844.28398	1332	1750	1508	9606	0	73.94	72.2	1.25
175	Q2	2.75	13379.67057	1286	1846	1451	8781	0	51.86	111.8	1.25

Thank you!

Note: For background, I am running this as part of a 2SLS analysis. This is
one of the reduced equations and I am hoping to extract values of Bhat to
feed back into the main structural equations. I am also having a similar
problem with the other reduced equation (they are part of a two-equation
system of simultaneous equations). 


--
View this message in context:
http://r.789695.n4.nabble.com/Coefficients-for-lagged-plm-model-variables-not-calculated-tp3879867p3879867.html
Sent from the R help mailing list archive at Nabble.com.

You probably should not do panel data analysis but multiple time series
analysis as your T is much larger than N. You only have seven units of
observation but some 150 observations on each unit.

Also, if the values on each unit of observation are very close to each other
between t and t-1, then you induce quasi-collinearity by including both in
your model. It could be that plm automatically kicks (quasi-) collinear
variables (without telling you).

Best,
Daniel

--
View this message in context:
http://r.789695.n4.nabble.com/Coefficients-for-lagged-plm-model-variables-not-calculated-tp3879867p3880390.html
Sent from the R help mailing list archive at Nabble.com.