summary() should not print anything

[koheiw]

Just like the older version of head.dfm(), summary.textmodel_wordfish_fitted() prints model parameters. It should only return values just like summary.lm().

ieWF <- dfm(data_corpus_irishbudget2010, removePunct = TRUE) %>%
    textmodel_wordfish(dir = c(6,5))
textplot_scale1d(ieWF)
out <- summary(ieWF)
# Call:
#   textmodel_wordfish.dfm(x = ., dir = c(6, 5))
# 
# Estimated document positions:
#   theta         SE       lower       upper
# 2010_BUDGET_01_Brian_Lenihan_FF        1.8209535 0.02032345  1.78111954  1.86078748
# 2010_BUDGET_02_Richard_Bruton_FG      -0.5932779 0.02818836 -0.64852708 -0.53802872
# 2010_BUDGET_03_Joan_Burton_LAB        -1.1136753 0.01540256 -1.14386429 -1.08348625
# 2010_BUDGET_04_Arthur_Morgan_SF       -0.1219325 0.02846319 -0.17772032 -0.06614462
# 2010_BUDGET_05_Brian_Cowen_FF          1.7724224 0.02364097  1.72608615  1.81875874
# 2010_BUDGET_06_Enda_Kenny_FG          -0.7145784 0.02650254 -0.76652337 -0.66263342
# 2010_BUDGET_07_Kieran_ODonnell_FG     -0.4844821 0.04171476 -0.56624299 -0.40272114
# 2010_BUDGET_08_Eamon_Gilmore_LAB      -0.5616713 0.02967351 -0.61983141 -0.50351124
# 2010_BUDGET_09_Michael_Higgins_LAB    -0.9703106 0.03850541 -1.04578121 -0.89484000
# 2010_BUDGET_10_Ruairi_Quinn_LAB       -0.9589229 0.03892373 -1.03521343 -0.88263242
# 2010_BUDGET_11_John_Gormley_Green      1.1807220 0.07221463  1.03918133  1.32226270
# 2010_BUDGET_12_Eamon_Ryan_Green        0.1866457 0.06294119  0.06328093  0.31001039
# 2010_BUDGET_13_Ciaran_Cuffe_Green      0.7421895 0.07245436  0.60017891  0.88420001
# 2010_BUDGET_14_Caoimhghin_OCaolain_SF -0.1840821 0.03666256 -0.25594076 -0.11222351

summary.textmodel_wordscores_fitted() is also wrong in the same way.

[kbenoit]

Should work like summary.lm().

Checklist:

• summary.textmodel_wordfish_fitted
• summary.textmodel_wordscores_fitted
• summary.textmodel_wordscores_predicted
• summary.textmodel_wordshoal_fitted

[koheiw]

I started working on in summary method for textmodel_wordscore(). Why does it use S4? To my mind, there is not reason to use S4 for text models, and it actually making the code unnecessarily complex.

[koheiw]

Fixing summary methods forces me to tidy up text models.

As far as we use generic predict(), text models' output should be consistent with other functions like lm(), but they are very different. predict.lm() returns only a named vector, unless se.fit = TRUE. Even so, its output is a plain list. From the lm examples:

> predict(lm.D9)

    1     2     3     4     5     6     7     8     9    10    11    12    13    14    15    16    17    18 
5.032 5.032 5.032 5.032 5.032 5.032 5.032 5.032 5.032 5.032 4.661 4.661 4.661 4.661 4.661 4.661 4.661 4.661 
   19    20 
4.661 4.661 

> predict(lm.D9, se.fit = TRUE)

$fit
    1     2     3     4     5     6     7     8     9    10    11    12    13    14    15    16    17    18 
5.032 5.032 5.032 5.032 5.032 5.032 5.032 5.032 5.032 5.032 4.661 4.661 4.661 4.661 4.661 4.661 4.661 4.661 
   19    20 
4.661 4.661 

$se.fit
 [1] 0.2202177 0.2202177 0.2202177 0.2202177 0.2202177 0.2202177 0.2202177 0.2202177 0.2202177 0.2202177
[11] 0.2202177 0.2202177 0.2202177 0.2202177 0.2202177 0.2202177 0.2202177 0.2202177 0.2202177 0.2202177

$df
[1] 18

$residual.scale
[1] 0.6963895

predict.textmodel_wordscore() returns data.frame always. From the textmodel_wordscre examples:

> predict(ws)

Predicted textmodel of type: wordscores

   textscore LBG se   ci lo   ci hi
R1   -1.3179 0.0067 -1.3311 -1.3048
R2   -0.7396 0.0114 -0.7620 -0.7172
R3    0.0000 0.0120 -0.0235  0.0235
R4    0.7396 0.0114  0.7172  0.7620
R5    1.3179 0.0067  1.3048  1.3311
V1   -0.4481 0.0119 -0.4714 -0.4247

If we make predict.textmodels_*() similar to predict.lm(), we can remove summary.textmodel_*_predicted and print.textmodel_*_predicted to make the code simpler. (we probably do not need the textmodel_*_predicted class). Removing those methods, nonetheless dose not cause users any inconvenience, because they can extract parameters by summary.textmodel_*_fitted() and see them nicely by print.textmodel_summary() (a new class that I am adding).