Stata not recognising binary variables as integers

[May Kingston]

Hello,

I have a binary variable (X1) which is already in ones and zeroes on an Excel spreadsheet and I'm trying to interact this with a continuous variable (X2) but every time I do I get the error message 'factor variables may not contain noninteger values' for the binary variable. I've tried the command to declare the variable - xtreg y i.x1#c.x2, fe but that's how I keep getting the error message. There's maybe one or two missing values but other than that there's nothing out of place in the dat - I wondered if it was something to do with missing values

Does anyone have any ideas on why this is happening or what I can do?

Thanks in advance.

[Andrew Musau]

One way to find out:

l X1 if X1!= int(X1), sep(0)

[Carlo Lazzaro]

May:
welcome to this forum.
I would investigate (and share on this forum thereafter) the -format- of your variable(s), as you can see in the following toy-example:

. use "C:\Program Files\Stata17\ado\base\a\auto.dta"
(1978 automobile data)

. format*

  Variable name  Display format
  -----------------------------
  make           %-18s
  price          %8.0gc
  mpg            %8.0g
  rep78          %8.0g
  -----------------------------
  headroom       %6.1f
  trunk          %8.0g
  weight         %8.0gc
  length         %8.0g
  turn           %8.0g
  -----------------------------
  displacement   %8.0g
  gear_ratio     %6.2f
  foreign        %8.0g
  -----------------------------

.

I addition, what you're experiencing has nothing to do with missing values, as Stata simply omit all observations with at least a missing value in any variables (listwise deletion).

[May Kingston]

Hello,

Thank you for the responses, I couldn't get your command to work Andrew

I used factor* and it says %10.0g which it what it says for the rest of them except for the id variable

[Andrew Musau]

I couldn't get your command to work Andrew

What does that mean? It did not display anything? Can you show exactly what you typed and what you got back?

[Nick Cox]

Please show results of

tab X1

or

tab x1

depending on which is the real name. Indeed, is that the problem, as you use both names in your post #1 and it’s entirely legal to have one of each?

[Maarten Buis]

So that indicates that your variable is stored in double precision. That is fine, it could still be an integer as far as the factor variable notation is involved. However, we know that that is not the case. Next question: are all values positive? With all values I mean all, that is without a single exception. So no missing values coded as -99.

[May Kingston]

x1/X1 is called ci2m I just called it x1 for simplicity. This is what happens when I use tab which is bizarre, it's almost like it's pulling the data from another variable because ci2 is purely '0' and '1'

tab ci2

ci2 | Freq. Percent Cum.
------------+-----------------------------------
0 | 130 10.23 10.23
.0604027 | 3 0.24 10.46
.1208054 | 1 0.08 10.54
.165697 | 758 59.64 70.18
.1812082 | 2 0.16 70.34
.2260998 | 16 1.26 71.60
.2416109 | 3 0.24 71.83
.2488163 | 12 0.94 72.78
.2834731 | 5 0.39 73.17
.2865025 | 3 0.24 73.41
.3020136 | 15 1.18 74.59
.309219 | 1 0.08 74.67
.3469052 | 3 0.24 74.90
.3696218 | 1 0.08 74.98
.4073079 | 4 0.31 75.30
.4145133 | 101 7.95 83.24
.4491701 | 3 0.24 83.48
.474916 | 5 0.39 83.87
.5095729 | 1 0.08 83.95
.5353188 | 2 0.16 84.11
.5926921 | 2 0.16 84.26
.5957215 | 2 0.16 84.42
.6303783 | 1 0.08 84.50
.6530948 | 3 0.24 84.74
.6561242 | 5 0.39 85.13
.690781 | 1 0.08 85.21
.6979864 | 26 2.05 87.25
.7134976 | 3 0.24 87.49
.7165269 | 10 0.79 88.28
.7511837 | 2 0.16 88.43
.7583891 | 3 0.24 88.67
.7739003 | 2 0.16 88.83
.8187919 | 5 0.39 89.22
.834303 | 5 0.39 89.61
.8791946 | 5 0.39 90.01
.9395973 | 9 0.71 90.72
1 | 118 9.28 100.00
------------+-----------------------------------
Total | 1,271 100.00

[May Kingston]

I think what's happened is Excel is rounding them up and down and displaying them as ones and zeroes but Stata is using the original values that it can somehow see but the data is supposed to be binary from the source - really weird

[Nick Cox]

Clearly Stata was right in reporting what it did. Perhaps the 0s and. 1s occur in the first 248 or so observations and the other values are in later observations and originated elsewhere. Such messes can arise in various ways, such as copy and paste.

[Nick Cox]

#10 crossed with #9. If Stata is so stupid that it can’t infer the intent you have from the content you supply, then you may need to give a push with the round() function.