For artificial data set 1, we know from the section "Bias due to between-source referral" that the appropriate log-linear model is model 11, and we see in Table 5 that this does indeed recover the true value accurately, estimating the missing cell count to be 85,500 (95% CI: 82,000, 89,100).
We showed in the section titled "Bias due to between-source referral" and in Appendix 1 that none of the 11 log-linear models corresponds to the truth for the model used to generate artificial data set 2. In fact, none of the 4 saturated models in Table 6 produces a 95% confidence interval that includes the true value.
Using the 3-source case as an example, we will show that between-source referrals will often correspond to a 3-way interaction term in the log-linear model.
However, the main source of referral is self-referral.
The main source of referral advice is the Referral Guidelines for Suspected Cancer (NICE, 2005).
The analysis examines four different sources of referrals: family referrals, social/legal agency referrals, school referrals and health/mental health referrals.
In multivariate analysis adjusting for gender, age, region of origin, HIV serostatus, and source of referral we identified source of referral as the major predictor for delays in LTBI therapy (Table 2).
Police are a major source of referral to psychiatric services but the appropriateness of these referrals has been questioned.
Source of referral was categorized into Immigration, Primary Medical Doctor, and Other (self-referrals and walk-ins).
My most common source of referral is from previous patients, and I really do value a patient who is informed when she comes in to the office.
Preference for the source of referral guidelines to be used in Europe.
