You might think you could simply use getElementsByTagName for this, but you can't. getElementsByTagName searches recursively and returns a single list for all the elements it finds. Since p elements can contain other p elements, you can't use getElementsByTagName, because it would return nested p elements that you don't want. To find only direct child elements, you'll need to do it yourself.

Example 10.16. Finding direct child elements

    def randomChildElement(self, node):
        choices = [e for e in node.childNodes
                   if e.nodeType == e.ELEMENT_NODE]   
        chosen = random.choice(choices)             
        return chosen                              

	As you saw in Example 9.9, “Getting child nodes”, the childNodes attribute returns a list of all the child nodes of an element.
	However, as you saw in Example 9.11, “Child nodes can be text”, the list returned by childNodes contains all different types of nodes, including text nodes. That's not what you're looking for here. You only want the children that are elements.
	Each node has a nodeType attribute, which can be ELEMENT_NODE, TEXT_NODE, COMMENT_NODE, or any number of other values. The complete list of possible values is in the __init__.py file in the xml.dom package. (See Section 9.2, “Packages” for more on packages.) But you're just interested in nodes that are elements, so you can filter the list to only include those nodes whose nodeType is ELEMENT_NODE.
	Once you have a list of actual elements, choosing a random one is easy. Python comes with a module called random which includes several useful functions. The random.choice function takes a list of any number of items and returns a random item. For example, if the ref elements contains several p elements, then choices would be a list of p elements, and chosen would end up being assigned exactly one of them, selected at random.