4
$\begingroup$

Many publications write about "completeness". I have a vague idea that it has to do with the ratio of how many sources from an observation are detected and how many sources are actually in that field. However, can someone please explain to me:

a) What is the definition of completeness?
b) How do you calculate the completeness of an observation?

As an example: Smolčić et al. 2008 are writing about completeness. Please have a look at A New Method to Separate Star-forming from AGN Galaxies at Intermediate Redshift: The Submillijansky Radio Population in the VLA-COSMOS Survey

$\endgroup$
1
  • 1
    $\begingroup$ I was going to ask for a block quote from where "completeness" appears in the linked paper, but it occurs 27 times! For example; "Section 4.1.2. Completeness and contamination due to the photometric selection" $\endgroup$
    – uhoh
    Jan 25 at 15:25
4
$\begingroup$

a) What is the definition of completeness?


Completeness is the number of objects in a data set that are detected over the number that exist. In astronomy, completeness is often estimated for a particular apparent magnitude or flux density. As an example, for sources that are as bright as the Sun (-27 magnitude), we have a completeness of 1. That is, we’ve discovered all sources in the celestial sphere with the Sun’s magnitude. As we get down to very dim objects like tiny pieces of space junk in Earth orbits, our completeness ratio drops dramatically.

b) How do you calculate the completeness of an observation?


On a ROC curve, completeness is the True Positive Rate or the y-axis if all the objects have been observed (thanks @ProfRob!). If only a fraction of the objects have been observed, completeness is that fraction times the true positive rate. Note that a classifier only yields better completeness at the cost of higher false positive rates. Here is an excellent example of ROC curves for different algorithms which differentiate quasars from stars:

The left panel shows data used in color-based photometric classification of stars and quasars. Stars are indicated by gray points, while quasars are indicated by black points. The right panel shows ROC curves for quasar identification based on u - g , g - r , r - i , and i - z colors.

enter image description here

If you don’t know the true positive rate, you may be able to run a simulation in which you inject objects into your data set and determine what portion of these are found by your algorithm. This will give you an estimate of your completeness. Matlab has a specific function for ROC curve Monte Carlo simulation, for example.

$\endgroup$
3
  • 1
    $\begingroup$ The true positive rate only tells you the completeness of your sample if you've observed all the objects. e.g. I work on cluster membership and can produce ROC curves. They do not tell me how complete my sample is. For that I need to know what fraction of objects were observed. $\endgroup$
    – ProfRob
    Jan 26 at 0:07
  • $\begingroup$ Thank you for your comment! I have a follow-up question. How does one simulate the data? I expect something like: 1. removing real sources from the data leaving just the noise and artifacts, 2. injecting fake sources. How does one find the correct distribution for 2. and does the source's morphology play a role? (if possible an answer in the radio astronomy context would be best). $\endgroup$
    – kelpfish
    Jan 26 at 7:11
  • $\begingroup$ @kelpfish You are quite welcome! In my opinion, your follow-up question exceeds the scope of the comment section. I would suggest posing it as a separate Astronomy Stack Exchange Question. $\endgroup$
    – Connor Garcia
    Jan 26 at 16:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.