jackknife_variance hardcoded for binary func

[smishr]

In jackknife.jl line 152,

θhj = func(design.data[!, x], rep_weights)

The func will only apply over a single data vector x. This is fine for mean and total, but wont work for ratio, which needs y

In general case, we need a type system here so that func with different number of args can work here

[codetalker7]

This issue is still relevant, but this PR changes the jackknife_variance function and calls it variance: #297

[ayushpatnaikgit]

We can do the following:

function variance(x::Vector{Symbol}, func::Function, design::ReplicateDesign{BootstrapReplicates})
    θ̂ = func(design.data, x, design.weights)
    θ̂t = [
        func(design.data, x, "replicate_"*string(i)) for
        i = 1:design.replicates
    ]
    variance = sum.((θ̂t .- θ̂) .^ 2) ./ design.replicates
    return DataFrame(estimator = θ̂, SE = sqrt(variance))
end

function ratio(df::DataFrame, columns, weights)
       return sum(df[!, columns[1]], StatsBase.weights(df[!, weights])) / sum(df[!, columns[2]], StatsBase.weights(df[!, weights]))
end

In the case of GLM,

function glm_tmp(df::DataFrame, columns, weights, link, family)
   formula_str = string("$(columns[1]) ~ ", join(columns[2:end], " + "))
   formula = eval(Meta.parse("@formula($formula_str)"))
   coef(glm(formula, data, link, family))
end

@nadiaenh can you try this?

[ayushpatnaikgit]

Can you also accept args, and kwargs in the variance functions and pass them to func? @nadiaenh