For asymptote estimations of species richness better to use samples. Individuals cpould be more useful in gradient analyses.

since bata have a tendency to aggregate and as it is secondary forest, it is better to do sample analysis rather than individual analysis

”Should I use individual-based or sample-based input for estimating bat species richness in a secondary forest?”

sample-based

Solitary or not, and with any kind of organisms, better base this type of biodiversity analysis (e.g., using rarefaction) on samples. 

Also consider using capture recapture-based estimators that explicitly account for not detecting all species, and the fact that species have different detection probabilities. You can use program CAPTURE, and select the Mh Jackknife estimator.

This question is methodologically difficult to answer, because some bats tends to live aggregated in roosts, but others live basically individually in the forest; and all feeds in different sites (i.e. different ecological niches). This pops out another questions, because the analisys depends heavily on the methods. ¿Are you sampling in roosts (i.e. caves, mines or something similar)? ¿Or are you sampling in the forest where the bats fly freely (and those that were aggregated will disperse) to seek for food? ¿Do you sample with mist nets, or harp traps, or bat ultrasound detectors, or a combination of them? If you use nets or traps, ¿are they only at ground level (0 to 3 m) or are they set also at the canopy level? If so, ¿how much capture effort at which level? ¿Are you focused on a single ensamble (let's say fruit bats or understory bats) or in the whole community (including, for example, the hard to capture high height flying molossids)? As you can see, to help you to answer your question, regarding on bats, is a little more complex tan expected...

The answer depends on your purpose and design. I suggest reading: Nicholas J. Gotelli and Robert K. Colwell. Quantifying biodiversity: procedures and pitfalls in

the measurement and comparison of species richness. Ecology Letters, (2001) 4: 379-391

I sampled in the forest (because the forest does not have any caves) where bats move freely but I can't help but think I accidentally located my traps near roosts (I spotted around three hovering fruit bats under a big tree where I put my mist net).  I was using harp nets, mist nets, and two kind of modified mist net installations (stacked vertically above without reaching canopy and covering the buildings of field stations). I finally used sample-based inputs for my rarefaction curve because of the strong spatial aspect (I guess).

This left me wondering one more thing: I also read in Maurer & McGill (the same book, different chapter 2011) that we can pool different methods to make rarefaction curves. I was wondering, is it really OK to pool bats captured by all those methods I had used in one rarefaction curves?

Bats do not seem an easy target…

You can combine the data from different sampling methods if they do not differ much in their efficiency (although each could be biased towards certain groups of species with common trait or habits).

At least theoretically, if you placed nets (of whatever design) at randomly selected locations, then you shall not worry about how near or close are the roosts. Later, after collecting sufficient data, you can test if the spatial distribution of caught bats are clumped (for roosting) or not and also check other possibilities. Do you use Ecosim for rarefactions, or other software? Just in case, here is the link to Ecosim:

http://www.garyentsminger.com/ecosim/ 

It also contains helpful tutorial and theory written in part by Nick Gotelli 

Actually, I think the trap was not really random because I locate them in a rather linear basis. I use EstimateS for rarefaction. I just use it to estimate possible additional species on my assemblage according several recommended estimators but I have not found yet how to test the spatial distribution using EstimateS. Thank you for the link and suggestion, I will try to search for possible spatial distribution.

EstimateS is a very good sofware for assessing number of species and also sample-based rarefaction. Ecosim is similar, but along with rarefactions (both individual- and sample -based) it also does various types of analyses including spatial distributions.

Placing traps on linear basis can still be considered close to random (hap-hazard), if the  line directions are not parallel but more or less random and the distance between traps are not regular.