Interpreting Coefficients of a Dummy variables derived from an Ordinal variable I have a variable that is measure societal complexity (SC) on a 3 point scale. 1 being the least complex and 3 being the most complex, and I think that this can safely be classed as a ordinal variable. 
However, the software for the model I am using this variable in currently does not allow ordinal variables, so I decided to convert this into 2 binary variables for SC1 and SC3, using SC2 as the baseline. 
My question is how should I intepret these coefficients? My intention when doing this was to be able to say SC1 is so much worse than SC2 and that SC3 is so much better than SC2. 
However, upon further contemplation, are these coefficients saying something closer to SC1 is so much worse than not SC1 (ie. SC2 & SC3) and that SC3 is so much better than not SC3 (ie. SC1 & SC2)? 
If so, is there a way to test something closer to my intention? 
Thanks.
EDIT
I am using an adaptation of a method developed by Freckleton & Jetz (2009) which controls for and estimates the influences of Phylogenetic and Spatial interdependencies in a linear model. The method is developed in R - although is not available in any CRAN package. Although I think my question applies to the general use of binary variables in a linear model. 
I have found little information discussing the matter. Although a number of texts have suggested converting likert scale scores into a number of binary variables (which can also arguabley be described as ordinal varibles), they do not discuss the issue of in-group vs out-group when using a number of dummy variables representing a ordinal variable. 
EDIT
Having considered recoding the data so that SC1 was the baseline, and then using an ordered approach that was suggested by Mark Step below and could be tabled as such:
   2.  3. 
1. 0   0
2. 1   1
3. 0   1

This would be equivalent to 
   1.  2.
1. 1   0
2. 0   1
3. 0   0

Which is the same result we started with, so I am unsure if this suggestion is any different to my intial attempt.
 A: You probably need a bit more detail in your question.
You can codify a factor with p-levels using indicator variables. The interpretation is unchanged. If you were to form a linear model then the parameter estimates are set based on the codes used for your indicator (binary) variables.
As a basic example, say you had constructed a model with a predictors cMWeight, I1 and I2 where I1 and I2 codified a 3-level factor (as in your case) and you get the following results:
Coefficients:
             Value Std. Error  t value Pr(>|t|)
(Intercept) 1.0312     0.0029 352.2184   0.0000
cMWeight    0.0503     0.0011  47.5039   0.0000
I1          0.0722     0.0042  17.3240   0.0000
I2         -0.0245     0.0041  -5.9097   0.0000

If you had coded SC0 as $I1=1$ and $I2=1$ then your model for that factor level would be formed by adopting those values in the regression equation, something along the lines of:
${\rm response} = 1.0312 + (0.0503 \times {\rm cMWeight}) + (0.0722 \times I_1) + ((-0.0245) \times I_2) + \epsilon$
Setting I1 and I2 to 1 gives your result for SC0. You can derive analogous results for each of your coded levels. In this example the interpretation results in just a different intercept term.
