"Bystander effect" means that if you measure the probability that someone will help a victim, you find a lower probability if there are other people present than if they are alone. The only time it makes sense to speak of a "bystander effect" is by testing that contrast: intervention probability alone vs. intervention probability with others present.
However, the Wikipedia page is not only talking about this specific "bystander effect", but also mixing in a broader collection of studies that measure probability of a bystander intervening. The header is "variables affecting bystanders"; the effect called "bystander effect" would be the specific case where the "variable affecting bystanders" is "presence of other people", but for a broader perspective the Wikipedia article is talking about other variables. For example, in one of the studies referenced:
Faul, M., Aikman, S. N., & Sasser, S. M. (2016). Bystander intervention prior to the arrival of emergency medical services: comparing assistance across types of medical emergencies. Prehospital Emergency Care, 20(3), 317-323.
they've fit a logistic regression model where the outcome (dependent variable) is "bystander intervention" and the predictor variables include gender, age, urbanicity, location, severity/condition (based on EMS's first impression of the case), and type of EMS response. They are not measuring 'bystander effect', and don't intend to. In fact, in some of the cases in their model, no bystander intervened because there was no bystander at all. They're looking more generally at reasons for bystander intervention/not.
I'm also not really sure where you got that phrase "depends on", as it isn't present in the Wikipedia page you link, but generally if you do encounter that phrase with respect to bystander effect I would take that to mean that the size/magnitude of the bystander effect depends on (item); this is dependence in the statistical/probability theory sense, in contrast to an effect being "independent" of some characteristic. There's no need for a moderating variable to be related to the primary manipulation of a study.
For example, let's say you measure bystander effect in a psychology laboratory by putting your participant in a room with one or more actors, and one of the actors drops some books they are carrying. You calculate bystander effect by measuring the rate at which your participant helps with the books when the only people in the room are the participant and single actor, or when there are 4 other actors present who do nothing about the books. Let's say you do this with a few dozen different participants and find that on average they are twice as likely to help when they are alone with the book-dropper.
Now, you repeat the experiment, except your participants spend 30 minutes browsing StackOverflow before they go into the room. In this case, you find that your participants are now three times as likely to help when they are alone with the book dropper. You could now say the bystander effect is dependent on browsing an internet Q&A site. You don't mean the bystander effect is caused by internet Q&A, and browsing an internet Q&A site doesn't have any relationship to "number of people in the room", but you've found a statistical dependence. You should probably do some other comparisons, though, too, to help understand what's going on: for example, was there a bigger effect because fewer people responded in a group, or because more people responded when they were alone with the book dropper, or parts of both? You might generate some more experiments to discover how specific this effect is and what parameters matter - is it specific to StackOverflow or do pictures of kittens have the same effect? Do you get the same result if participants complete a math worksheet or programming exercise? Etc..