When you are in SIM mode, you stay on the same m/z (or a few m/z) longer than when you scan the full spectrum. Therefore the signal-to-noise ratio is much better for the ion of interest. Selectivity come from the choice of the m/z value(s), which should be specific to the compound under scrutiny.

In scan mode using Quadrupole mass analyzer, ions are scanned one after one. At the given moment, only ions with one m/z (e.g., 18) reach the detector (electron multiplier), other ions (17, 19 , etc.) have other trajectories and do not reach the detector. When other ions (not your target) are detected, your ions are discarded and do not reach the detector. Therefore, only a small fraction of your target ions actually reaches the detector.  In SIM mode when a single ion is monitored (best case) all ions having target m/z value reach the detector. The difference in sensitivity depends on a mass range of scan, number of ions monitored in SIM and the relative abundance of the monitored ion (highest sensitivity is achieved for compounds mass spectra of which have one abundant ion). Typical difference is 1-2 orders of magnitude. For Ion trap or TOF MS, the difference will be lower because no ions are lost.

I absolutely agree with the above answers BUT it depends what you are trying to do.  If you have an unknown sample which you are trying to characterize then you need to use SCAN mode so you can look at the full range of compounds present and identify them via their mass spectra.  If however you are looking at specific molecules or groups of molecules (for example the alkane fraction of a bituminous sample) then you would use SIM mode.  The only drawback then is that you may have to run the sample more than once (once in SCAN and once in SIM), or use a GC-MS which allows you to do both at the same time, to identify which alkanes are present and their individual retention times.  Because the SIM mode is specific to the compounds you are looking for,  it may allow them to be seen when they might not be easily identifiable in SCAN mode and it certainly gives more reliable quantification because you are not losing ions so get a more representative result.

Typically, one uses scan for qualitative identification and SIM for quantitative analysis.  Qualitative scans allow you to pick your quant and qual ion(s).  The reason SIM is so much for sensitive has to do with data acquisition electronics.  For each second, your acquisition card can only collect so many data points.  If you are scanning across a mass range, say 50 - 500 amu at 1 amu resolution, you must divide the total data points by your range, i.e. 500 in my example above.  The more data points you can collect at each amu, the more S/N averaging you typically achieve.  Hence, if you then go to SIM mode and only have 3 amu, you are now able to collect 167X as many data points than in scan at each amu.  These data points are averaged therefore your S/N is significantly higher.  If you have fast scan electronics, you can sample even faster and stand much less chance of missing peak information in fast GC/MS runs (i.e. using H2 carrier gas) while maintaining the minimum >20 data points across the peak.

the above said is all very true. - I just want to add one more thought. the described increase in sensitivy can only be reached if the limit of determination is governed by physical noise of the machine, than the increased time gives increased sensitivity, if the LOQ is determined by chemical noise (dirt in the ion source) you will not experience any increased sensitivity by using SIM. - This is the main reason why in HPLC-MS not SIM is used but multireaction monitoring.

Yes,all are correct. When everything is known and you wanted only to do quantitative analysis, it is best to use sIM. Because of those electronic configurations cited above. But if you are uncertain yet of your analyses, of couse you have to use Scan first.