Combining discrete variable and continuous variables in a linear model framework I'm trying to assess the influence of temperature, geography and larvae connectivity on the genetic structure (Fst values) of a marine species.
I used a linear regression framework, which examine the correlation between the genetic variation (FST) and temperature/geography (GEO) values. 
For instance giving the following command in R:
lm(Fst ~ GEO + TEMP, data)

On the other side, I have larvae connectivity (CO) values that I want to include in the model I build.
For instance the model would be:
lm(Fst ~ GEO + TEMP + CO)

However, the connectivity values are either 0 (not connected) or 1 (connected).
These values follow more a logistic regression than a linear one. 
Then, how could I include these connectivity values test in my linear model with temperature and geography?
When I tested lm(Fst ~ CO), the residuals seems correct.
Could I trust the results and use a simple linear regression model for these data?
Thanks for your help.
 A: There's no issue whatever in having a 0/1 variable on the right hand side of your model (i.e. as a predictor/"independent variable"/covariate). Binary variables and factors more generally are common in regression models. 
That doesn't make it "logistic"; you might use a logistic model if you had a 0/1 variable on the left hand side of the model (as a response/"dependent variable").
[As to whether you can "trust the results", that depends on other things than whether there's a 0/1 variable in your model.]
A: In the second model, when you apply ordinary linear squares (OLS) regression, the binary variable CO is dummy coded: $X$ will either take the value $0$ or $1$. 
From the point of view of linear algebra, your second model will be as follows:
$$\Tiny \begin {bmatrix}  
Fst_1\\ Fst_2\\Fst_3\\\vdots\\ Fst_n
\end {bmatrix}  = \Tiny \begin {bmatrix}  
1&GEO_1&TEMP_1&CO_1\\ 1&GEO_2&TEMP2&CO_2\\1&GEO_3&TEMP_3&CO_3\\1&\vdots&\vdots&\vdots\\ 1&GEO_n&TEMP_n&CO_n
\end {bmatrix} \small \begin {bmatrix}  
\beta_0\\ \beta_1\\\beta_2\\ \beta_n
\end {bmatrix} $$
... Just like any other system. But while all the other variable values in the model matrix (i.e. $\small GEO$ and $\small TEMP$ are numeric, continuous realizations of random variables with typically a wide range of real values, $\small CO$ will be a bunch of zeros and ones depending on whether each particular observation or subject (remember that observations or subjects are points in the data cloud, and correspond to individual rows in the model matrix... I guess specimens of the marine species you are studying) is "connected" ($X=1$, or in our case, $\small CO_i=1$) or "not connected" ($\small CO_i=0$), whatever that means in your dataset. Effectively, then, $\beta_n$ acts like a switch (when $\small CO_i=0$, the switch is off, and we don't add any contribution from the coefficient of $\small CO$ (i.e. $\small \beta_n$); when $\small CO_i=1$ the effect of $\small \beta_n$ is added to the prediction).
Therefore, when $X=1$, we are just adding a constant to the intercept (or subtracting) - in effect, the intercept will be $\beta_0+\beta_n$, instead of just $\beta_0$.
As @Glen_b was explaining, in logistic regression the idea is completely unrelated: we are modelling a system where a number of regressors are trying to explain the odds of something happening or not happening, a binary outcome, on the left-hand side of the equation. On the LHS we have the $ln\,(odds)$ and on the RHS we have a system pretty much as above with or without dummy variables. We assume that the outcome ($Y=1$) follows a binomial distribution with probability $pi$, and come up with the logit function $ln(\frac{\pi}{1-\pi})$ for the LHS.
