NOTE: This program by Cliff Clogg is an old workhorse but users today are much better served by the prpgram LEM written by Jeroen Vermunt. There is a reference to LEM on the "Links to Other Latent Class Resources" page. The LEM program and reference manual with lots of examples can be downloaded free from that link.

Program Set-Up for MLLSA

MLLSA is an old-fashioned card-oriented program that has been adapted from a main-frame version to run on PCs. Input to the program is from a flat text (ascii) file with each line (card) in a specified format. MLLSA is executed by a call of the form:

MLLSA <in-file> <out-file>

where <in-file> is the file containing program instructions following the format described below and <out-file> is the file where output from the program is placed.

Card Status Columns Format Description
1. Title N 1-80 20A4 Alphameric title
2. Problem N 1-2 I2 # manifest variables
  N 3-4 I2 # latent classes
  OP 5-10 F6.0 Sample size (check value)
  OP 11-16 I6 Max iterations (default=500)
  OP 17-24 F8.7 Max deviation (default=.00005)
  OP 25-26 I2 Enter 1 for chi-sq test of independence
  OP 27-28 I2 Enter 1 for assignment of respondents to latent classes
  OP 29-30 I2 Blank
  OP 31-32 I2 Enter 1 if variables labels are input (card 4)
  OP 33-34 I2 Enter 1 if values labels are input (card 5)
  OP 35-36 I2 Enter 1 if LC probability restrictions (card 10)
  OP 37-38 I2 Enter 1 if conditional probability restrictions (card 11)
  OP 39-40 I2 Enter 1 if iteration detail is desired (1st and last iterations)
  OP 41-42 I2 Enter 1 to output standardized residuals
  OP 43-44 I2 Enter 1 if start values are final values from previous problem
  OP 45-46 I2 Enter 1 to output rank of information matrix
  OP 48 I1 # of groups (31-34 must be blank; sample size must be given)
3. Levels N 1-80 40I2 # categories (levels) for manifest variables in 2-digit fields
4. Labels OP 1-80 10(4X,A4) Alphameric variable names (5-8, 13-16, etc.)
5. Values OP 1-80 10(4X,A4) Value labels (5-8, 13-16, etc.); ONE CARD PER VARIABLE
6. Format N 1-80 20A4 Data format [e.g., (8F4.0)]
7. Input Data N 1-80   Frequencies in above format; see NOTE 1
8. LC Start Values N 1-80 10F8.7 Initial values for LC proportions; must sum to 1; multiple cards as necessary
9. CP Start Values N 1-80 10F8.7 Initial values for conditional probabilities; see NOTE 2
10. LC Restrictions OP 1-80 20I2 Restrictions on LC proportions; see NOTE 3
11. CP Restrictions OP 1-80 20I4 Restrictions on conditional probabilities; see NOTE 4

NOTE 1 - Frequencies are entered for cells with subscripts for the first variable varying most rapidly. E.g., for dichotomous variables A, B, C, D the order of cells would be (1111), (2111), (1211), (2211), (1121),...,(2222). If there is a grouping variable, it must be the last variable.

NOTE 2 - Input one card (or set of cards if necessary) per manifest variable with start values for conditional probabilities for first LC, second LC, etc. E.g., for 3 latent classes and 4 dichotomous manifest variable, the start values might be:

000000.2000000.8000000.5000000.5000000.7000000.3
000000.2000000.8000000.5000000.5000000.7000000.3
000000.2000000.8000000.5000000.5000000.7000000.3
000000.2000000.8000000.5000000.5000000.7000000.3

Note that the start values must sum to 1 across the categories of each manifest variable.

NOTE 3 - Enter only if columns 35-36 of Problem Card are not blank. Enter 0 (or leave blank) for fields corresponding to free parameters; enter 1 for fields with parameters that are fixed to start values; enter integers greater than 1 in other fields where any two or more integers equal to each other will constrain these parameters to be equal. E.g., for a 5-class model where it is desired to constrain the first LC probability to equal the start value, to leave the second LC probability free and to constrain the final 3 LC probabilities to be equal, the appropriate card would be, starting in column 1: 0100020202

NOTE 4 - Enter only if columns 37-38 of Problem Card are not blank. Enter 0 (or leave blank) for fields corresponding to free parameters; enter 1 for fields with parameters that are fixed to start values; enter integers greater than 1 in other fields where any two or more integers equal to each other will constrain these parameters to be equal. The structure (not format) of these cards is the same as in NOTE 2, above. Restrictions may apply within or between manifest variables.