R语言中创建data.frame新列时`df$NewCol=`与`df[,"NewCol"]=`的区别是什么？为何POSIXlt类型赋值时二者表现不同？

Great questions—let's break this down clearly, using concrete examples where it matters.

1. Differences between df$NewCol= and df[,"NewCol"]= for adding columns #

These two syntaxes might seem interchangeable at first, but they have key behavioral differences:

• Column name matching: The $ operator supports partial matching of column names. For example, if your data frame has a column named NewColumn, df$New will automatically match it (as long as the partial name is unique). On the other hand, df[,"New"] requires an exact match by default—you'd need to add exact=FALSE to enable partial matching here.
• Handling non-atomic objects: This ties directly to your second question. The $ operator will accept almost any object as a new column (including lists or S3 class objects like POSIXlt), storing it as a list column if needed. The [ assignment, however, tries to coerce the right-hand side into an atomic vector (the standard type for data frame columns), which can lead to unexpected behavior or warnings if the object isn't an atomic vector.
• Subset return behavior (for reading, not just writing): While less relevant for assignment, it's worth noting that df$NewCol always returns a vector, whereas df[,"NewCol"] returns a vector by default but will return a single-column data frame if you use df[,"NewCol", drop=FALSE].

2. Why as.POSIXlt causes a warning with [ but not $ in the iris example #

Let's start with what as.POSIXlt(Sys.Date()) actually returns: it's an S3 object of class POSIXlt, which is under the hood a list with 9 components (like sec, min, hour, mday, etc.). Data frames typically expect columns to be atomic vectors (like character, integer, or POSIXct—which is a numeric vector representing seconds since epoch), not lists.

Here's what happens with each assignment:

• When you run iris$NewCol = as.POSIXlt(Sys.Date()), the $ operator doesn't overthink it—it just takes the entire POSIXlt list object and stores it as a list column in the data frame. No coercion is attempted, so no warnings are thrown, and the assignment works as intended.
• When you run iris[,'NewCol'] = as.POSIXlt(Sys.Date()), the [ assignment logic tries to fit the right-hand side into a standard atomic vector column. Since POSIXlt is a list with 9 components, R misinterprets this as you trying to assign 9 separate values/variables to a single column. This triggers the warning about "provided 9 variables to replace 1 variable"—and what actually ends up in the column is just the first component of the POSIXlt list (all zeros, in this case, since Sys.Date() sets time to midnight).

For basic atomic types (character, integer, float), both syntaxes work the same because those are exactly the kind of atomic vectors that [ expects—no coercion issues arise, so no warnings.

内容的提问来源于stack exchange，提问作者explorer